Magnetic Resonance Texture Analysis in Alzheimer's disease

  • Author Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
    Jia-Hui Cai
    Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China
    Search for articles by this author
  • Author Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
    Yuan He
    Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China
    Search for articles by this author
  • Author Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
    Xiao-Lin Zhong
    Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
    Affiliations
    Institute of Clinical Medicine, The First Affiliated Hospital of University of South China, Hengyang, Hunan, China
    Search for articles by this author
  • Hao Lei
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China
    Search for articles by this author
  • Fang Wang
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China
    Search for articles by this author
  • Guang-Hua Luo
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China
    Search for articles by this author
  • Heng Zhao
    Correspondence
    Address correspondence to: H.Z.
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China

    Department of Radiology, Shengjing Hospital of China Medical University, Sanhao Street No. 36, Heping District, Shenyang 110004, China
    Search for articles by this author
  • Jin-Cai Liu
    Affiliations
    Department of Radiology, The First Affiliated Hospital of University of South China, Chuanshan Road No. 69, Hengyang 421000, Hunan, China
    Search for articles by this author
  • Author Footnotes
    1 Jia-Hui Cai, Yuan He and Xiao-Lin Zhong contributed equally to this paper.
Open AccessPublished:February 10, 2020DOI:https://doi.org/10.1016/j.acra.2020.01.006
      Texture analysis is an emerging field that allows mathematical detection of changes in MRI signals that are not visible among image pixels. Alzheimer's disease, a progressive neurodegenerative disease, is the most common cause of dementia. Recently, multiple texture analysis studies in patients with Alzheimer's disease have been performed. This review summarizes the main contributors to Alzheimer's disease-associated cognitive decline, presents a brief overview of texture analysis, followed by review of various MR imaging texture analysis applications in Alzheimer's disease. We also discuss the current challenges for widespread clinical utilization. MR texture analysis could potentially be applied to develop neuroimaging biomarkers for use in Alzheimer's disease clinical trials and diagnosis.

      Key Words

      Abbreviations:

      AD (Alzheimer's disease), MCI (mild cognitive impairment), MRTA (Magnetic resonance texture analysis), TA (texture analysis), NFTs (neurofibrillary tangles), (amyloid-β), ROS (reactive oxygen species), VBM (voxel-based morphometry), CSF (cerebrospinal fluid), ROI (region of interest), AUC (area under the curve), GLCM (gray-level co-occurrence matrix), GLRLM (gray-level run-length matrix), LBP (Local binary patterns), 2D (two-dimensional), 3D (three-dimensional), HOC (hippocampal occupancy), LBD (Lewy body dementias), LBP-TOP (LBP three orthogonal planes)

      INTRODUCTION

      Alzheimer's disease (AD) is a progressive neurodegenerative disease characterized by memory loss and multiple cognitive impairments [
      • Reddy P.H.
      • Oliver D.M.
      Amyloid beta and phosphorylated tau-induced defective autophagy and mitophagy in Alzheimer's Disease.
      ]. Globally, over 47 million people live with dementia, and this number is expected to triple by 2050 [
      • Livingston G.
      • Sommerlad A.
      • Orgeta V.
      • et al.
      Dementia prevention, intervention, and care.
      ]. The global cost of dementia was estimated to be US$818 billion in 2015, and the cost will continue to increase as the number of people with dementia rises [
      • Livingston G.
      • Sommerlad A.
      • Orgeta V.
      • et al.
      Dementia prevention, intervention, and care.
      ]. The majority of dementia cases are AD-related. Given these facts, it is important to learn to understand, treat, and prevent AD. Although significant attention has been paid to the treatment of AD, little progress has been made [
      • Liu X.
      • Chen K.
      • Wu T.
      • et al.
      Use of multimodality imaging and artificial intelligence for diagnosis and prognosis of early stages of Alzheimer's disease.
      ]. The disease has no specific cure; the condition of the patient worsens with the disease progression, eventually leading to death. However, early therapeutic interventions in patients with AD could potentially delay the progression from mild cognitive impairment (MCI) and thus reduce costs associated with long-term care [
      • Eskildsen S.F.
      • Coupe P.
      • Garcia-Lorenzo D.
      • et al.
      Prediction of Alzheimer's disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning.
      ]. In addition, the annual rates of MCI conversion to dementia are as high as 20% [
      • Langa K.M.
      • Levine D.A.
      The diagnosis and management of mild cognitive impairment: a clinical review.
      ], and accurate predictions may reduce the costs associated with selecting a pharmaceutical trials subject when conducting large-scale trials of disease modifying drugs [
      • Eskildsen S.F.
      • Coupe P.
      • Garcia-Lorenzo D.
      • et al.
      Prediction of Alzheimer's disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning.
      ]. Therefore, methods to improve the ability to diagnose and predict AD are needed.
      In both the recommendations by International Working Group (known as the IWG criteria) [
      • Dubois B.
      • Feldman H.H.
      • Jacova C.
      • et al.
      Advancing research diagnostic criteria for Alzheimer's disease: the IWG-2 criteria.
      ] and the National Institute of Neurological Disorders and Stroke-Alzheimer Disease and Related Disorders (known as NINCDS-ADRDA criteria) [
      • Dubois B.
      • Feldman H.H.
      • Jacova C.
      • et al.
      Research criteria for the diagnosis of Alzheimer's disease: revising the NINCDS-ADRDA criteria.
      ], the use of neuroimaging for diagnosis and prognosis of AD has been highlighted. Neuroimaging biomarkers play an important role in advancing our understanding of AD. Advanced neuroimaging methods such as diffusion imaging, resting-state functional MRI, and voxel-based morphometry (VBM) are less helpful in detecting the unseen signal changes. Magnetic resonance texture analysis (MRTA), an application of radiomics, can detect subtle signal fluctuation and obtain latent image information. The major hallmarks of early-stage AD, such as neurofibrillary tangles (NFTs) and amyloid-β (Aβ) peptide, cannot be detected with the resolution of current clinical MRI, but their accumulated effects on the brain tissue cause changes in MRI image pixel intensity. These changes can form certain texture patterns in the MRI images, which may be captured by texture analysis (TA) [
      • Sorensen L.
      • Igel C.
      • Liv Hansen N.
      • et al.
      Early detection of Alzheimer's disease using MRI hippocampal texture.
      ,
      • Castellano G.
      • Bonilha L.
      • Li L.M.
      • et al.
      Texture analysis of medical images.
      ]. This review focuses on the role of MRTA in AD. Here we review the main contributors to AD-associated cognitive decline, the basic concepts behind MRTA, MRTA applications in AD, current challenges, and potential future directions.

      MAIN CONTRIBUTORS TO AD-ASSOCIATED COGNITIVE DECLINE

      AD is mainly characterized by extracellular aggregated amyloid fibrils and plaques (mostly composed of fibrillary Aβ peptide) and intracellular NFTs (mainly built up of hyperphosphorylated tau protein), leading to dysfunction and loss of synapses and eventual neuronal death [
      • Butterfield D.A.
      • Halliwell B.
      Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease.
      ,
      • Martins R.N.
      • Villemagne V.
      • Sohrabi H.R.
      • et al.
      Alzheimer's disease: a journey from amyloid peptides and oxidative stress, to biomarker technologies and disease prevention strategies-gains from AIBL and DIAN Cohort Studies.
      ].

       Oxidative Stress and Mitochondrial Defects

      Oxidative stress, the result of an imbalance between the production of reactive nitrogen species and reactive oxygen species (ROS), and intracellular antioxidant defense [
      • Butterfield D.A.
      • Halliwell B.
      Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease.
      ], has been shown to contribute to the pathological progress of AD in a wide range of studies [
      • Butterfield D.A.
      • Halliwell B.
      Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease.
      ,
      • Martins R.N.
      • Villemagne V.
      • Sohrabi H.R.
      • et al.
      Alzheimer's disease: a journey from amyloid peptides and oxidative stress, to biomarker technologies and disease prevention strategies-gains from AIBL and DIAN Cohort Studies.
      ,
      • Butterfield D.A.
      • Di Domenico F.
      • Barone E.
      Elevated risk of type 2 diabetes for development of Alzheimer disease: a key role for oxidative stress in brain.
      ,
      • Cheignon C.
      • Tomas M.
      • Bonnefont-Rousselot D.
      • Faller P.
      • Hureau C.
      • Collin F.
      Oxidative stress and the amyloid beta peptide in Alzheimer's disease.
      ,
      • Butterfield D.A.
      • Boyd-Kimball D.
      Oxidative stress, amyloid-beta peptide, and altered key molecular pathways in the pathogenesis and progression of Alzheimer's Disease.
      ,
      • Di Domenico F.
      • Pupo G.
      • Giraldo E.
      • et al.
      Oxidative signature of cerebrospinal fluid from mild cognitive impairment and Alzheimer disease patients.
      ,
      • Di Domenico F.
      • Tramutola A.
      • Butterfield D.A.
      Role of 4-hydroxy-2-nonenal (HNE) in the pathogenesis of alzheimer disease and other selected age-related neurodegenerative disorders.
      ]. ROS such as inducible nitric oxide synthase, hydroxyl radicals, and cyclooxygenase-2 cause oxidative damage to the deoxyribonucleic acid (DNA), proteins, and membrane lipids [
      • Li R.
      • Zhang Y.
      • Rasool S.
      • et al.
      Effects and underlying mechanisms of bioactive compounds on type 2 diabetes mellitus and Alzheimer's disease.
      ,
      • Nita M.
      • Grzybowski A.
      The Role of the reactive oxygen species and oxidative stress in the pathomechanism of the age-related ocular diseases and other pathologies of the anterior and posterior eye segments in adults.
      ]. When ROS are produced in excess, they directly affect mitochondria [
      • Lejri I.
      • Agapouda A.
      • Grimm A.
      • et al.
      Mitochondria- and oxidative stress-targeting substances in cognitive decline-related disorders: from molecular mechanisms to clinical evidence.
      ], which module brain cell survival and death by producing ATP and endogenous ROS, maintaining calcium homeostasis and redox equilibrium, and controlling cell apoptosis [
      • Lejri I.
      • Agapouda A.
      • Grimm A.
      • et al.
      Mitochondria- and oxidative stress-targeting substances in cognitive decline-related disorders: from molecular mechanisms to clinical evidence.
      ,
      • Yan M.H.
      • Wang X.
      • Zhu X.
      Mitochondrial defects and oxidative stress in Alzheimer disease and Parkinson disease.
      ]. The presence of Aβ peptide and NFTs causes increases in ROS and mitochondrial dysfunction. Meanwhile, mitochondrial dysfunction leads to increased ROS, which then enhances Aβ peptide aggregation, tau hyperphosphorylation, and NFT formation [
      • Ahmad W.
      • Ijaz B.
      • Shabbiri K.
      • Ahmed F.
      • Rehman S.
      Oxidative toxicity in diabetes and Alzheimer's disease: mechanisms behind ROS/ RNS generation.
      ]. All these effects accelerate the progression of AD.

       Neuroinflammation

      Inflammation is critical for initiating tissue repair, maintaining basal cognitive and homeostatic functions; however, sustained neuroinflammation has injurious effects on neurological function, disrupting cognition, and promoting neurodegeneration [
      • Bettcher B.M.
      • Kramer J.H.
      Inflammation and clinical presentation in neurodegenerative disease: a volatile relationship.
      ,
      • Brown K.L.
      • Cosseau C.
      • Gardy J.L.
      • et al.
      Complexities of targeting innate immunity to treat infection.
      ]. Inflammatory cytopathology includes microglial and astrocytic reactions [
      • Bostanciklioglu M.
      An update on the interactions between Alzheimer's disease, autophagy and inflammation.
      ]. Microglia, the resident macrophage population of the central nervous system, play a major role in this neuroinflammation [
      • McGeer E.G.
      • McGeer P.L.
      Neuroinflammation in Alzheimer's disease and mild cognitive impairment: a field in its infancy.
      ,
      • Ginhoux F.
      • Lim S.
      • Hoeffel G.
      • et al.
      Origin and differentiation of microglia.
      ]. Aβ peptide within the central nervous system [
      • Subhramanyam C.S.
      • Wang C.
      • Hu Q.
      • et al.
      Microglia-mediated neuroinflammation in neurodegenerative diseases.
      ] and increased expression of inducible nitric oxide synthase [
      • Bajwa E.
      • Pointer C.B.
      • Klegeris A.
      The role of mitochondrial damage-associated molecular patterns in chronic neuroinflammation.
      ] induce microglial activation, initiating a proinflammatory cascade that results in the release of proinflammatory cytokines, leading to surrounding neuronal injury and cell death [
      • Schwab C.
      • McGeer P.L.
      Inflammatory aspects of Alzheimer disease and other neurodegenerative disorders.
      ]. Meanwhile, prolonged neuroinflammation exacerbates Aβ burden and tau hyperphosphorylation. In this manner, neuroinflammation leads to progressive tissue damage.

       Brain Insulin Resistance

      Brain insulin resistance is defined as the marked attenuation of the ability of insulin to elicit a response within brain cells [
      • Mielke J.G.
      • Taghibiglou C.
      • Liu L.
      • et al.
      A biochemical and functional characterization of diet-induced brain insulin resistance.
      ]. This lack of response may be due to downregulation of the insulin receptor, an incapability of insulin receptor to bind insulin, or faulty activation of the insulin signaling cascade [
      • Arnold S.E.
      • Arvanitakis Z.
      • Macauley-Rambach S.L.
      • et al.
      Brain insulin resistance in type 2 diabetes and Alzheimer disease: concepts and conundrums.
      ]. The binding of insulin to its receptors inhibits cell apoptosis by inhibiting glycogen synthase kinase 3β activity [
      • Desai G.S.
      • Zheng C.
      • Geetha T.
      • et al.
      The pancreas-brain axis: insight into disrupted mechanisms associating type 2 diabetes and Alzheimer's disease.
      ]. Aβ-peptide [
      • Rodriguez-Casado A.
      • Toledano-Diaz A.
      • Toledano A.
      Defective insulin signalling, mediated by inflammation, connects obesity to alzheimer disease; relevant pharmacological therapies and preventive dietary interventions.
      ] and activation of proinflammatory pathways [
      • De Felice F.G.
      • Ferreira S.T.
      Inflammation, defective insulin signaling, and mitochondrial dysfunction as common molecular denominators connecting type 2 diabetes to Alzheimer disease.
      ] lead to impaired insulin signaling, and subsequent increased glycogen synthase kinase 3β activity, hyperphosphorylation of tau, formation of NFTs, and increased production of Aβ peptide [
      • Li R.
      • Zhang Y.
      • Rasool S.
      • et al.
      Effects and underlying mechanisms of bioactive compounds on type 2 diabetes mellitus and Alzheimer's disease.
      ,
      • Phiel C.J.
      • Wilson C.A.
      • Lee V.M.
      • et al.
      GSK-3alpha regulates production of Alzheimer's disease amyloid-beta peptides.
      ], which then impair neuronal activity and regeneration, neurite retraction, and synapse formation [
      • Tumminia A.
      • Vinciguerra F.
      • Parisi M.
      • et al.
      Type 2 diabetes mellitus and Alzheimer's disease: role of insulin signalling and therapeutic implications.
      ,
      • Li L.
      • Holscher C.
      Common pathological processes in Alzheimer disease and type 2 diabetes: a review.
      ].

      TEXTURE ANALYSIS

      TA is an emerging methodology that allows mathematical detection of changes in MRI signals that are not visible among image pixels, thereby providing a quantitative and reproducible method of extracting image features [
      • Zhang Y.
      • Zhu H.
      • Mitchell J.R.
      • et al.
      T2 MRI texture analysis is a sensitive measure of tissue injury and recovery resulting from acute inflammatory lesions in multiple sclerosis.
      ,
      • de Carvalho Alegro M.
      • Valotta Silva A.
      • Yumi Bando S.
      • et al.
      Texture analysis of high resolution MRI allows discrimination between febrile and afebrile initial precipitating injury in mesial temporal sclerosis.
      ]. The image texture generally refer to the spatial variation in pixel intensity levels within a tissue [
      • Alobaidli S.
      • McQuaid S.
      • South C.
      • et al.
      The role of texture analysis in imaging as an outcome predictor and potential tool in radiotherapy treatment planning.
      ]. Changes in image intensity due to the deposition of Aβ peptide and NFTs may be reflected as certain textural patterns before neuronal death (Fig 1). Figure 2 illustrates a simplified workflow for the clinical implementation of TA.
      Figure 1
      Figure 1Schematic view of the proposed texture working hypothesis in AD. Top row: NFTs inside the neurons and Aβ plaques between neurons spread throughout the brain, causing neuronal death. Bottom row: changes in the statistical properties of the image intensities due to the accumulated effect of NFTs and/or Aβ plaques may be reflected as certain textural patterns prior to atrophy. (Color version of figure is available online.)
      Figure 2
      Figure 2Simplified model of MRTA workflow. MRTA, magnetic resonance texture analysis.

       Types of TA

      TA methods can be categorized as statistical, structural, model-based, and transform-based, according to the approach used to evaluate the inter-relationships of the pixels [
      • Castellano G.
      • Bonilha L.
      • Li L.M.
      • et al.
      Texture analysis of medical images.
      ].
      Statistical-based TA uses properties controlling the distribution and relationships of gray-level values in the image to represent texture [
      • Castellano G.
      • Bonilha L.
      • Li L.M.
      • et al.
      Texture analysis of medical images.
      ]. First-order statistical TA, also known as a histogram, extracts the image intensity values within the region of interest (ROI). A histogram can be generated by calculating a frequency count of the number of pixels of each gray-level intensity value [
      • Chitalia R.D.
      • Kontos D.
      Role of texture analysis in breast MRI as a cancer biomarker: a review.
      ]; from the resulting histogram, many parameters may be derived. Table 1 defines these histogram features.
      Table 1Histogram (First Order) Texture Features
      Histogram FeaturesQualitative Description
      MeanThe average value of the pixels within the region of interest
      Standard deviation (SD)Measure of how much variation or dispersion exists from the average (mean value)
      SkewnessMeasures the asymmetry of the histogram of pixel intensities within a region of interest
      KurtosisIndicates how tall and sharp the central peak is relative to the normal distribution curve
      EntropyHISTRefer to the number of different pixel intensities within a region of interest. Entropy is therefore a measure of disorder
      EnergyHISTRefer to the uniformity of an image
      Mean of positive pixels (MPP)Average of the pixels that have positive pixel intensities. Positive pixels are pixels that are brighter than the mean
      Second-order statistical methods analyze the spatial relationship or co-occurrence of the pixel intensity values. Several methods exist to analyze second-order statistics, the two most common are gray-level co-occurrence matrix (GLCM) and gray-level run-length matrix (GLRLM) methods. The GLCM analyze the gray-level distribution of pairs of pixels in a specified distance and a specified image orientation [
      • Chitalia R.D.
      • Kontos D.
      Role of texture analysis in breast MRI as a cancer biomarker: a review.
      ], reflecting texture/spatial arrangements of pixel intensities present in the ROI [
      • Dragic M.
      • Zaric M.
      • Mitrovic N.
      • et al.
      Application of gray level co-occurrence matrix analysis as a new method for enzyme histochemistry quantification.
      ]. Two-dimensional (2D) GLCM is usually quantified in 4 directions (0°, 45°, 90°, and 135°) and three-dimensional (3D) GLCM in 13 directions [
      • Soni N.
      • Priya S.
      • Bathla G.
      Texture analysis in cerebral gliomas: a review of the literature.
      ]. GLCM features are further defined in Table 2. GLRLM estimates the spatial relationships between groups of pixels with similar gray-level values; run-length features allows evaluate the coarseness of texture in a predetermined direction [
      • Kassner A.
      • Thornhill R.E.
      Texture analysis: a review of neurologic MR imaging applications.
      ]. GLRLM features are described in Table 3. GLCM and GLRLM provide fine texture in short distance and run, and provide coarse texture in longer distance and run [
      • Soni N.
      • Priya S.
      • Bathla G.
      Texture analysis in cerebral gliomas: a review of the literature.
      ]. The coarseness of texture is related to spatial frequency. Fine texture with high spatial frequency, whereas coarse texture with low spatial frequency. The fine (≤2 mm), medium (3–5 mm), and coarse texture (>6 mm) can be extracted using different filter values.
      Table 2Gray-level Co-occurrence Matrix Texture Features
      GLCM
      GLCM, Gray-level Co-occurrence matrix.
      Features
      Qualitative Description
      EntropyGLCMMeasures disorder of pixel intensity relationships within a region of interest
      EnergyGLCMMeasures uniformity of pixel intensity relationships within a region of interest
      ContrastMeasures the quantity of local variations within pixel intensity relationships within an image
      CorrelationMeasures a potential connection between a pixel and its local neighborhood of pixels, reflecting the image gray level correlation
      Inverse different momentMeasures the smoothness (homogeneity) of the gray level distribution of the image
      HomogeneityMeasure the closeness of distribution in the co-occurrence matrix to the matrix diagonal
      Cluster shadeMeasure asymmetry in gray-level values
      low asterisk GLCM, Gray-level Co-occurrence matrix.
      Table 3Gray-Level Run-length Matrix Texture Features
      GLRLM
      GLRLM, Gray-level Run-length Co-occurrence matrix.
      Features
      Qualitative Description
      Short run emphasis (SRE)Emphasis on short runs
      Long run emphasis (LRE)Emphasis on long runs
      Gray-level nonuniformity (GLN)Degree of gray-level run dissimilarity
      Run length nonuniformity (RLN)Dissimilarity in length of runs
      Run percentage (RP)Distribution of runs
      Low gray-level run emphasis (LGRE)Emphasis on low gray-level values
      High gray-level run emphasis (HGRE)Emphasis on high gray-level values
      High gray-level run emphasis (HGRE)Emphasis on short runs with low gray-level values
      Short run high gray-level emphasis (SRHGE)Emphasis on short runs with high gray-level values
      low asterisk GLRLM, Gray-level Run-length Co-occurrence matrix.
      Structural-based TA captures changes in intensity between the central voxel and surrounding neighboring voxels [
      • Chitalia R.D.
      • Kontos D.
      Role of texture analysis in breast MRI as a cancer biomarker: a review.
      ]. Local binary patterns (LBP) is the most common method. LBP is a nonparametric algorithm that describes the local features of the gray-level relationship between an image pixel and each surrounding neighboring pixel; a generalized gray-scale and rotation invariant operator and multiresolution analysis are also proposed for TA [
      • Sorensen L.
      • Shaker S.B.
      • de Bruijne M.
      Quantitative analysis of pulmonary emphysema using local binary patterns.
      ].
      Model-based TA interprets image texture with generative image model, such as stochastic and fractal models. The features of the model are extracted and then used for image analysis. This method lacks direction selectivity and is not suitable for describing local image structures.
      Transform-based TA converts the spatial information of an image to spatial frequencies of that image. Transform-based methods include wavelet, Fourier, and Gabor transforms [
      • Chitalia R.D.
      • Kontos D.
      Role of texture analysis in breast MRI as a cancer biomarker: a review.
      ]. Wavelet transform, the most widely used method, analyze the frequency content of the image within different spatial-frequency resolutions [
      • Castellano G.
      • Bonilha L.
      • Li L.M.
      • et al.
      Texture analysis of medical images.
      ].

       Machine Learning

      There is increasing interest in the use of artificial intelligence in medical imaging. Machine learning is a branch of artificial intelligence that enables computers to learn from existing “training data” and creates complex analytical models [
      • Reig B.
      • Heacock L.
      • Geras K.J.
      • et al.
      Machine learning in breast MRI.
      ]. Firstly, this technique can be used for feature selection. Another use is assisting clinical diagnostic decision in classifying and predicting AD [
      • Ahmed M.R.
      • Zhang Y.
      • Feng Z.
      • et al.
      Neuroimaging and machine learning for dementia diagnosis: recent advancements and future prospects.
      ]. Machine learning applications for medical imaging are condensed into three broad paradigms: supervised learning, unsupervised learning, and semisupervised learning. Supervised learning aims to identify links between features related to the learning target and desired outcome measures in the dataset to achieve a classification decision. Unsupervised learning aims to recognize and determine latent patterns using the unlabeled data provided by computer. Semisupervised learning is a combination of supervised and unsupervised machine learning that make use of a large amount of unlabeled data and a small amount of labeled data for training [
      • Handelman G.S.
      • Kok H.K.
      • Chandra R.V.
      • et al.
      eDoctor: machine learning and the future of medicine.
      ].

      TEXTURE ANALYSIS APPLICATIONS IN AD

      MRTA applications in AD are a promising research field, and multiple reports have shown encouraging results (Table 4, Table 5, Table 6). We have divided the various studies into three broad categories: MRTA for classifying AD, predicting AD, differentiating AD from other types of dementia.
      Table 4MRTA in Classification of AD
      AuthorNo. of PatientsMagnet Strength/MRI SequenceTissue/Structure(s)Type of Texture AnalysisClassificationClassification MethodsResults
      Simoes R et al1361.5T/3D-T

      1WI
      Whole-brainLBP-TOP
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      NC
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      -AD
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      SVM
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      LBP-TOP performs accurate classification and localizes discriminative brain regions
      Feng F et al1163.0T/3D-T1WIHippocampusHistogram

      GLCM
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      Wavelet

      GLRLM
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      NC-AD

      AD-aMCI
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      aMCI-NC
      SVMHippocampus has significantly different texture features among AD and aMCI patients and NC
      De Oliveira M.S et al492.0T/3D-T1WICorpus Callosum

      Thalamus
      GLCMNC-AD

      AD-aMCI

      aMCI-NC
      TA
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      parameters showed differences among AD-NC, aMCI-AD, and aMCI-NC for the Corpus Callosum and thalamus
      Zhang J et al343.0T/3D-T1WIhippocampus

      entorhinal cortex
      Histogram

      Absolute gradient

      Run-length matrix

      Co-occurrence matrix
      NC-ADANN
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      1-NN
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      Texture analysis could characterize the differences of texture features in the tissues in AD and NC
      Gao N et al2991.5T/3D-T1WIHippocampusContourlet-based transform

      GLCM
      NC-ADGR regression

      PLS algorithm
      The model containing texture features resulted in better classification performance than the one excluding texture features
      Sorensen L et al8351.5T, 3.0T/3D-T1WIHippocampusHistogramNC-AD

      NC-MCI
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      Logistic regressionCombining texture and volume to the model performs better than texture and volume alone in classification effects
      Luk C.C et al7901.5T/3D volumetric T1WIwhole-brainVGLCM-TOP-3D
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      NC-ADBinary logistic regressionA classification model incorporating texture features and HOC
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      performs better than texture and HOC alone in classification effects
      Oppedal K et al1821.0T,1.5T/FLAIR,3D-T1WIWML
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      WM
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      NAWM
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      VLBP
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      LBP-TOP
      1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.


      Contrast
      NC - ADRandom ForestThe two-class problems NC vs AD shows different texture features
      low asterisk 1-NN, 1-nearest neighbor; HOC, hippocampal occupancy; aMCI, amnestic mild cognitive impairment; AD, Alzheimer's disease; ANN, artificial neural network; AUC, area under the curve; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run-length matrix; LBP, local binary patterns; LBP-TOP, LBP three orthogonal planes; NAWM, normal appearing white matter; NC, normal control; SVM, support vector machine; TA, texture analysis; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis; VLBP, volume LBP; WML, white matter lesions; WM, all of white matter.
      Table 5MRTA in Prediction of AD
      AuthorNo. of PatientsMagnet Strength/MRI SequenceTissue/Structure(s)Type of Texture AnalysisPredictionConversion PeriodModeling MethodsResults (MCI-AD)
      Gao N et al2991.5 T/T1WIHippocampusContourlet-based transform

      GLCM
      AD, Alzheimer's disease; GLCM, gray-level co-occurrence matrix; HOC, hippocampal occupancy; MCI, mild cognitive impairment; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis.
      MCI
      AD, Alzheimer's disease; GLCM, gray-level co-occurrence matrix; HOC, hippocampal occupancy; MCI, mild cognitive impairment; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis.
      -AD
      AD, Alzheimer's disease; GLCM, gray-level co-occurrence matrix; HOC, hippocampal occupancy; MCI, mild cognitive impairment; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis.


      MCI stable

      MCI-NC
      24 moGR regression

      PLS algorithm
      Combining texture features to develop models can effectively improve the prediction of conversion from MCI
      Sorensen L et al8351.5 T, 3.0 T/3D-T1WIHippocampusHistogramMCI-AD24 moLogistic regressionHippocampal texture performs better than volume in prediction effect, and the predictive ability was improved with a model
      Luk C.C et al7901.5 T/3D volumetric T1WIWhole-brainVGLCM-TOP-3D
      AD, Alzheimer's disease; GLCM, gray-level co-occurrence matrix; HOC, hippocampal occupancy; MCI, mild cognitive impairment; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis.
      MCI-nonconverters

      MCI-converters
      36 moBinary logistic regressionThe ability of each texture feature to predict MCI to AD was higher than HOC
      AD, Alzheimer's disease; GLCM, gray-level co-occurrence matrix; HOC, hippocampal occupancy; MCI, mild cognitive impairment; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis.
      , and the predictive ability was improved with a model
      low asterisk AD, Alzheimer's disease; GLCM, gray-level co-occurrence matrix; HOC, hippocampal occupancy; MCI, mild cognitive impairment; VGLCM-TOP-3D, 3-dimensional voxel-based texture analysis.
      Table 6MRTA in the Differential Diagnosis of AD
      AuthorNo. of PatientsMagnet Strength/MRI SequenceTissue/Structure(s)Type of Texture AnalysisClassificationClassification MethodsResults
      Oppedal K et al1821.0 T, 1.5 T/FLAIR,3D-T1WML
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.


      WM
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.
      LBP
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.


      contrast
      AD
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.
      -LBD
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.
      Random forestThe two-class problem AD vs LBD shows different texture features
      Oppedal K et al1821.0 T, 1.5 T/FLAIR,3D-T1WML
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.


      WM

      NAWM
      VLBP
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.


      LBP-TOP

      contrast
      AD-LBDRandom forestThe two-class problem AD vs LBD shows different texture features
      Kodama N and Y Kawase71Cerebral parenchymaCo-occurrence matrix

      run-length matrix
      AD-DLB
      AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.
      Discriminant analysisThe study confirmed DLB and AD from patients with dementia, respectively
      low asterisk AD, Alzheimer's disease; DLB, dementia with Lewy bodies; LBD, Lewy body dementias; LBP, Local binary patterns; NAWM, normal appearing white matter; VLBP, volume LBP; WML, white matter lesions; WM, all of normal white matter.

       Texture Analysis in the Classification of AD

      Table 4 summarizes prior studies for classification of AD [
      • Sorensen L.
      • Igel C.
      • Liv Hansen N.
      • et al.
      Early detection of Alzheimer's disease using MRI hippocampal texture.
      ,
      • Simoes R.
      • van Cappellen van Walsum A.M.
      • Slump C.H.
      Classification and localization of early-stage Alzheimer's disease in magnetic resonance images using a patch-based classifier ensemble.
      ,
      • Feng F.
      • Wang P.
      • Zhao K.
      • et al.
      Radiomic features of hippocampal subregions in Alzheimer's disease and amnestic mild cognitive impairment.
      ,
      • de Oliveira M.S.
      • Balthazar M.L.
      • D'Abreu A.
      • et al.
      MR imaging texture analysis of the corpus callosum and thalamus in amnestic mild cognitive impairment and mild Alzheimer disease.
      ,
      • Zhang J.
      • Yu C.
      • Jiang G.
      • et al.
      3D texture analysis on MRI images of Alzheimer's disease.
      ,
      • Gao N.
      • Tao L.X.
      • Huang J.
      • et al.
      Contourlet-based hippocampal magnetic resonance imaging texture features for multivariant classification and prediction of Alzheimer's disease.
      ,
      • Luk C.C.
      • Ishaque A.
      • Khan M.
      • et al.
      Alzheimer's disease: 3-dimensional MRI texture for prediction of conversion from mild cognitive impairment.
      ,
      • Oppedal K.
      • Engan K.
      • Eftestøl T.
      • et al.
      Classifying Alzheimer's disease, Lewy body dementia, and normal controls using 3D texture analysis in magnetic resonance images.
      ]. Some of these are briefly discussed below. Because 3D TA has high spatial resolution, sensitivity and specificity than 2D techniques, most studies used 3D TA [
      • Zhang J.
      • Yu C.
      • Jiang G.
      • et al.
      3D texture analysis on MRI images of Alzheimer's disease.
      ,
      • Kovalev V.A.
      • Kruggel F.
      • Gertz H.J.
      • et al.
      Three-dimensional texture analysis of MRI brain datasets.
      ]. Simoes et al [
      • Simoes R.
      • van Cappellen van Walsum A.M.
      • Slump C.H.
      Classification and localization of early-stage Alzheimer's disease in magnetic resonance images using a patch-based classifier ensemble.
      ] performed 3D TA using LBP three orthogonal planes (LBP-TOP) computed at local image patches in the whole brain and found discrimination between normal control (NC) and AD patients groups. They further located brain regions in early AD, and found several patches are concentrated near hippocampus, with the left side being slightly more discriminative than the right side. But other studies have showed that the right side alterations are more apparent. Feng et al [
      • Feng F.
      • Wang P.
      • Zhao K.
      • et al.
      Radiomic features of hippocampal subregions in Alzheimer's disease and amnestic mild cognitive impairment.
      ] performed TA in 116 patients by using histogram, GLCM, GLRLM, and wavelet transformation, and discovered that the most significant changes were identified in the right hippocampus in the AD group. Similar changes were reported by De Oliveira et al [
      • de Oliveira M.S.
      • Balthazar M.L.
      • D'Abreu A.
      • et al.
      MR imaging texture analysis of the corpus callosum and thalamus in amnestic mild cognitive impairment and mild Alzheimer disease.
      ], who performed GLCM-based TA in 49 patients using T1-weighted images; they found the texture features for the left and right sides of the thalamus were substantially different among groups, and more significant features were found on the right side. The studies mentioned above indicate that asymmetric texture between the left and right sides may be an early sign of AD progression. A mechanism underlying asymmetric texture in AD may be metabolic asymmetry [
      • Fletcher P.T.
      • Powell S.
      • Foster N.L.
      • et al.
      Quantifying metabolic asymmetry modulo structure in Alzheimer's disease.
      ].
      Increasing the scope of TA to include cerebrospinal fluid (CSF) may improve classification accuracy. Zhang et al [
      • Zhang J.
      • Yu C.
      • Jiang G.
      • et al.
      3D texture analysis on MRI images of Alzheimer's disease.
      ] examined 3D TA of T1WI images in 34 patients and chose four texture features after feature selection and reduction that were significantly correlated with Mini-Mental State Examination scores. They found that a relatively large ROI containing part of the CSF produces a higher classification accuracy than an ROI only including the hippocampus and the entorhinal cortex. This may be because NFT and/or Aβ peptide deposition in the CSF creates more distinctive 3D texture features for extraction. However, due to the small sample size, the further studies should be investigated, even though the preliminary results of this study are encouraging.
      Machine learning techniques applied to the TA workflow have been shown to aid the classification of AD in some studies. Gao et al [
      • Gao N.
      • Tao L.X.
      • Huang J.
      • et al.
      Contourlet-based hippocampal magnetic resonance imaging texture features for multivariant classification and prediction of Alzheimer's disease.
      ] used contourlet transform and GLCM to extract 14 texture features (1344 parameters) from hippocampal areas in 299 subjects, and found that the model containing texture features resulted in better classification performance than the one excluding texture features. However, the study did not compare the classification performance of TA with that of the model. Sorensen et al [
      • Sorensen L.
      • Igel C.
      • Liv Hansen N.
      • et al.
      Early detection of Alzheimer's disease using MRI hippocampal texture.
      ] performed histogram-based TA using T1WI images and found combining texture and hippocampal volume using logistic regression produced an area under the curve (AUC) of 0.915 for discriminating NC from AD and an AUC of 0.806 for discriminating NC from MCI, which is higher than texture (AUC = 0.912 and 0.764) and volume (AUC = 0.909 and 0.784) alone; they also proved that TA is better than volume for this purpose. The hippocampal occupancy (HOC), which is calculated by hippocampal volume divided by hippocampal volume plus inferior lateral ventricle volume, performs better on discriminative accuracy compare to the standard hippocampal volume measure [
      • Heister D.
      • Brewer J.B.
      • Magda S.
      • et al.
      Predicting MCI outcome with clinically available MRI and CSF biomarkers.
      ]. Therefore, combining TA and HOC results in better performance. Luk et al [
      • Luk C.C.
      • Ishaque A.
      • Khan M.
      • et al.
      Alzheimer's disease: 3-dimensional MRI texture for prediction of conversion from mild cognitive impairment.
      ] used 3D voxel-based TA to calculate 8 texture features in 790 subjects. Differences in MRI texture changes were seen in all 8 texture features in a comparative analysis of NC and AD patient groups (AUC = 0.928), and a classification model incorporating all texture features and HOC produced an AU of 0.930, which was superior to HOC (0.843) and texture (0.928) alone.

       Texture Analysis in the Prediction of AD

      Table 5 summarizes MRTA studies on the prediction of conversion from MCI to AD [
      • Sorensen L.
      • Igel C.
      • Liv Hansen N.
      • et al.
      Early detection of Alzheimer's disease using MRI hippocampal texture.
      ,
      • Gao N.
      • Tao L.X.
      • Huang J.
      • et al.
      Contourlet-based hippocampal magnetic resonance imaging texture features for multivariant classification and prediction of Alzheimer's disease.
      ,
      • Luk C.C.
      • Ishaque A.
      • Khan M.
      • et al.
      Alzheimer's disease: 3-dimensional MRI texture for prediction of conversion from mild cognitive impairment.
      ]. Combining texture features and machine learning techniques can help improving predictive effects. Gao et al [
      • Gao N.
      • Tao L.X.
      • Huang J.
      • et al.
      Contourlet-based hippocampal magnetic resonance imaging texture features for multivariant classification and prediction of Alzheimer's disease.
      ] discovered that combining texture features to develop models using machine learning can effectively improve the prediction of conversion from MCI over a 2-year observation period. Furthermore, comparing the different modeling methods showed that the machine learning method heterogeneity influences predictive performance.
      Some studies used volume [
      • Tabatabaei-Jafari H.
      • Shaw M.E.
      • Walsh E.
      • et al.
      Cognitive/functional measures predict Alzheimer's disease, dependent on hippocampal volume.
      ], HOC [
      • Heister D.
      • Brewer J.B.
      • Magda S.
      • et al.
      Predicting MCI outcome with clinically available MRI and CSF biomarkers.
      ], and VBM [
      • Ferreira L.K.
      • Diniz B.S.
      • Forlenza O.V.
      • et al.
      Neurostructural predictors of Alzheimer's disease: a meta-analysis of VBM studies.
      ] as predictor, but the MRTA features outperformed these methods. Sorensen et al [
      • Sorensen L.
      • Igel C.
      • Liv Hansen N.
      • et al.
      Early detection of Alzheimer's disease using MRI hippocampal texture.
      ] noted that hippocampal texture achieved a significantly higher AUC in predicting MCI-to-AD conversion within 24 months compared to hippocampal volume; this finding supports that texture to some degree captures more information than volume, and that predictive ability is improved when adding MRI texture features to model. Similar evidence was reported by Luk et al [
      • Luk C.C.
      • Ishaque A.
      • Khan M.
      • et al.
      Alzheimer's disease: 3-dimensional MRI texture for prediction of conversion from mild cognitive impairment.
      ], who found that the performance of texture features to predict MCI converters (MCI-AD) and MCI nonconverters was higher than HOC alone within 36 months, and combining texture features and HOC into model using logistic regression yielded the greatest AUC. They also reported that hippocampus and parahippocampal gyrus texture changes can be seen in both left and right hemispheres, which differs from the results of previous VBM studies. The results of their meta-analysis [
      • Ferreira L.K.
      • Diniz B.S.
      • Forlenza O.V.
      • et al.
      Neurostructural predictors of Alzheimer's disease: a meta-analysis of VBM studies.
      ] showed that MCI converters have significant cluster of gray matter volumetric reduction in the left hemisphere, which may be due to the results from brain hypometabolism. This suggests that texture changes due to the deposition of Aβ peptide and NFTs may precede the development of atrophy caused by brain cell death.

       Texture Analysis in the Differential Diagnosis of AD

      Dementia with Lewy bodies and Parkinson disease dementia together comprise the Lewy body dementias (LBD) [
      • Hansen D.
      • Ling H.
      • Lashley T.
      • et al.
      Review: clinical, neuropathological and genetic features of Lewy body dementias.
      ]. LBD are mainly characterized by abnormal aggregation of α-synuclein. In addition, Aβ peptide and tau protein could also accumulate in the brain of patients with LBD. It is important to distinguish between AD and LBD, since they differ in prognosis and response to drug treatment. However, AD and LBD have overlapping clinical symptoms and pathological findings, making them challenging to diagnose, delay, or cure. Dopamine transporter scanning is an effective way to distinguish between AD and LBD, but it is expensive and not available at all centers.
      MRTA shows promising results in distinguishing AD from LBD (Table 6) [
      • Oppedal K.
      • Engan K.
      • Eftestøl T.
      • et al.
      Classifying Alzheimer's disease, Lewy body dementia, and normal controls using 3D texture analysis in magnetic resonance images.
      ,
      • Oppedal K.
      • Eftestol T.
      • Engan K.
      • et al.
      Classifying dementia using local binary patterns from different regions in magnetic resonance images.
      ,
      • Kodama N.
      • Kawase Y.
      Computerized method for classification between dementia with Lewy bodies and Alzheimer's disease by use of texture analysis on brain MRI.
      ]. Oppedal et al [
      • Oppedal K.
      • Engan K.
      • Eftestøl T.
      • et al.
      Classifying Alzheimer's disease, Lewy body dementia, and normal controls using 3D texture analysis in magnetic resonance images.
      ] used volume LBP, LBP-TOP and contrast values extracted from FLAIR and T1WI images in areas with normal white matter (segmented from the T1MR image), white matter lesions (localized areas of white matter hyperintensities on T2-weighted MR images and FLAIR images, segmented from the FLAIR images), and normal appearing white matter (calculated as the difference between white matter hyperintensities and normal white matter) to discern LBD and AD. They achieved a total accuracy in discriminating AD and LBD of 0.79 (0.15). Compared to the results they achieved in another study [
      • Oppedal K.
      • Eftestol T.
      • Engan K.
      • et al.
      Classifying dementia using local binary patterns from different regions in magnetic resonance images.
      ] using 2D LBP to differentiate AD from LBD, the results with 3D LBP were more accurate. They also found more texture information in the 3D-T1 image compared to the FLAIR image. Kodama and Kawase [
      • Kodama N.
      • Kawase Y.
      Computerized method for classification between dementia with Lewy bodies and Alzheimer's disease by use of texture analysis on brain MRI.
      ] calculated texture features from co-occurrence matrix and run-length matrix, and confirmed dementia with Lewy bodies in 7 of 10 patients (70.0%) with dementia and AD in 33 of 36 patients (91.7%) with the disease.

      CHALLENGES OF TEXTURE ANALYSIS

      Despite these advantages, widespread clinical application of MRTA are still challenging, mainly due to the lack of standardization. The image intensities discretization, acquisition parameters, MRTA software and methodologies all contributes to MRTA heterogeneity.
      The method of image intensity discretization, which entails resampling image intensity values, has an important effect on texture features. Texture features calculation requires separating image intensities into an appropriate number of discrete resampled values or bins to analyze an image. Discrete resampled values or bins are class interval used to divide pixel intensity data. There are two ways to achieve discretization: using a fixed number of discrete resampled values or a fixed intensity resolution (bin size). A fixed number of discrete resampled values require dividing the image intensity range into equally spaced intervals, where the intensity resolutions varies per image. This method is not suitable for texture feature values comparison between patients and within patients. A fixed intensity resolution in units of image intensity maintains a constant intensity resolution across all images. It is unknown whether intensity resolutions or spaced intervals are more important [
      • Patel N.
      • Henry A.
      • Scarsbrook A.
      The value of MR textural analysis in prostate cancer.
      ,
      • Leijenaar R.T.
      • Nalbantov G.
      • Carvalho S.
      • et al.
      The effect of SUV discretization in quantitative FDG-PET Radiomics: the need for standardized methodology in tumor texture analysis.
      ].
      The other important barrier to MRTA implementation relates to acquisition parameters. For example, Ford et al [
      • Ford J.
      • Dogan N.
      • Young L.
      • Yang F.
      Quantitative radiomics: impact of pulse sequence parameter selection on MRI-based textural features of the brain.
      ] found that texture features generated by the different pulse sequences vary considerably among T1-weighted images (spin echo, spoiled gradient echo, inversion recovery gradient echo, gradient echo, and inversion recovery spin echo) and the difference are also measured on the T1 map. Another phantom-based study performed by Waugh et al [
      • Waugh S.A.
      • Lerski R.A.
      • Bidaut L.
      • et al.
      The influence of field strength and different clinical breast MRI protocols on the outcome of texture analysis using foam phantoms.
      ] demonstrated that the variation of sequence parameters has less effect on the TA, whereas spatial resolution may have significant influence on MRTA. Solomon at al [
      • Solomon J.
      • Mileto A.
      • Nelson R.C.
      • et al.
      Quantitative features of liver lesions, lung nodules, and renal stones at multi-detector row CT examinations: dependency on radiation dose and reconstruction algorithm.
      ] performed a retrospective analysis by extracting 23 quantitative imaging features (texture, shape, attenuation, size, pixel value distribution, and edge sharpness), and found that reconstruction algorithms affect the extraction and analysis of texture features.
      The available evidence suggests that 3D TA, which contains more spatial information, outperforms 2D TA. The extension of 2D gray-scale methods to 3D need to consider neighboring voxels on the 3-D voxel lattice, as well as properties of 3D textures such as spatial anisotropy [
      • Kovalev V.A.
      • Kruggel F.
      • Gertz H.J.
      • et al.
      Three-dimensional texture analysis of MRI brain datasets.
      ].
      MRTA software may be commercial, open-source, or developed in-house. Although there is no evidence that any one type of software is superior, the different types of software capture different numbers of texture features, which may have an impact on results. There is also no standardization of feature extracting methodology and modeling methods, which limits the comparison of results across different studies.

      FUTURE DEVELOPMENTS

      Although several studies describe the role of TA in medical imaging, prospective multicenter studies are needed before translation to clinical implementation. To maximize the validity of future research, all centers should follow strict protocols and standardized methodologies.
      The combination of MRTA and CSF markers, one of the current diagnostic indicators for AD, increases our understanding of the connection between texture change and progression of AD [
      • Sorensen L.
      • Igel C.
      • Liv Hansen N.
      • et al.
      Early detection of Alzheimer's disease using MRI hippocampal texture.
      ]. In addition, MRTA combined with other quantitative imaging data, such as quantitative susceptibility maps, may improve early AD diagnosis and prognosis.
      Further, the texture features can be fused with deep learning for AD classification and prediction. The deep learning can learn pixel data directly, bypassing manual features selection and input, thus, the performance of the deep learning technologies can be better than that of ordinary classifiers [
      • Suzuki K.
      Overview of deep learning in medical imaging.
      ], and combining TA and deep learning may be helpful in classifying and predicting AD.
      The combination of radiomic features with genomic data is known as radiogenomics [
      • Gillies R.J.
      • Kinahan P.E.
      • Hricak H.
      Radiomics: images are more than pictures, they are data.
      ]. Mathys et al [
      • Mathys H.
      • Davila-Velderrain J.
      • Peng Z.
      • et al.
      Single-cell transcriptomic analysis of Alzheimer's disease.
      ] found that neuron and glial cell genes affect neuronal survival, axonal integrity, and myelination by AD pathology. In the future, combinations of radiomic and genomic features may be used to classify and predict the progression of AD and to further explore the relationship between specific texture features and underlying histopathology.

      Conclusion

      Currently, the presumptive diagnosis of AD is based on the combination of CSF markers and clinical phenotype. We are entering a new era of combined imaging and clinical evaluation that will increasingly depend on neuroimaging markers in disease detection and diagnosis. As reviewed here, emerging evidence supports the potential of TA as a neuroimaging marker of AD. MRTA provides an early, noninvasive method to detect AD, since it can detect the unseen signal changes. At present, although its implementation for AD imaging faces some challenges, MRTA shows promise for the diagnosis and prognosis of AD.

      Acknowledgments

      This work was supported by the Hunan Provincial Natural Science Foundation of China (grant numbers 2017JJ2225 and 2018JJ2357 ), Hunan Provincial Science and Technology Innovation Foundation of China (grant number 2017SK50203 ), Hunan Provincial Innovation Foundation For Postgraduate (grant number CX20190767 ), Scientific Research Key Project of Hunan Provincial Health Commission (grant number 20201911 ). Hunan Provincial Innovation Foundation For Postgraduate (grant number CX20190767 ).

      REFERENCES

        • Reddy P.H.
        • Oliver D.M.
        Amyloid beta and phosphorylated tau-induced defective autophagy and mitophagy in Alzheimer's Disease.
        Cells. 2019; 8: 1-21
        • Livingston G.
        • Sommerlad A.
        • Orgeta V.
        • et al.
        Dementia prevention, intervention, and care.
        Lancet (London, England). 2017; 390: 2673-2734
        • Liu X.
        • Chen K.
        • Wu T.
        • et al.
        Use of multimodality imaging and artificial intelligence for diagnosis and prognosis of early stages of Alzheimer's disease.
        Transl Res. 2018; 194: 56-67
        • Eskildsen S.F.
        • Coupe P.
        • Garcia-Lorenzo D.
        • et al.
        Prediction of Alzheimer's disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning.
        Neuroimage. 2013; 65: 511-521
        • Langa K.M.
        • Levine D.A.
        The diagnosis and management of mild cognitive impairment: a clinical review.
        JAMA. 2014; 312: 2551-2561
        • Dubois B.
        • Feldman H.H.
        • Jacova C.
        • et al.
        Advancing research diagnostic criteria for Alzheimer's disease: the IWG-2 criteria.
        Lancet Neurol. 2014; 13: 614-629
        • Dubois B.
        • Feldman H.H.
        • Jacova C.
        • et al.
        Research criteria for the diagnosis of Alzheimer's disease: revising the NINCDS-ADRDA criteria.
        Lancet Neurol. 2007; 6: 734-746
        • Sorensen L.
        • Igel C.
        • Liv Hansen N.
        • et al.
        Early detection of Alzheimer's disease using MRI hippocampal texture.
        Hum Brain Mapp. 2016; 37: 1148-1161
        • Castellano G.
        • Bonilha L.
        • Li L.M.
        • et al.
        Texture analysis of medical images.
        Clin Radiol. 2004; 59: 1061-1069
        • Butterfield D.A.
        • Halliwell B.
        Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease.
        Nat Rev Neurosci. 2019; 20: 148-160
        • Martins R.N.
        • Villemagne V.
        • Sohrabi H.R.
        • et al.
        Alzheimer's disease: a journey from amyloid peptides and oxidative stress, to biomarker technologies and disease prevention strategies-gains from AIBL and DIAN Cohort Studies.
        J Alzheimers Dis. 2018; 62: 965-992
        • Butterfield D.A.
        • Di Domenico F.
        • Barone E.
        Elevated risk of type 2 diabetes for development of Alzheimer disease: a key role for oxidative stress in brain.
        Biochim Biophys Acta. 2014; 1842: 1693-1706
        • Cheignon C.
        • Tomas M.
        • Bonnefont-Rousselot D.
        • Faller P.
        • Hureau C.
        • Collin F.
        Oxidative stress and the amyloid beta peptide in Alzheimer's disease.
        Redox Biol. 2018; 14: 450-464
        • Butterfield D.A.
        • Boyd-Kimball D.
        Oxidative stress, amyloid-beta peptide, and altered key molecular pathways in the pathogenesis and progression of Alzheimer's Disease.
        J Alzheimers Dis. 2018; 62: 1345-1367
        • Di Domenico F.
        • Pupo G.
        • Giraldo E.
        • et al.
        Oxidative signature of cerebrospinal fluid from mild cognitive impairment and Alzheimer disease patients.
        Free Radical Biol Med. 2016; 91: 1-9
        • Di Domenico F.
        • Tramutola A.
        • Butterfield D.A.
        Role of 4-hydroxy-2-nonenal (HNE) in the pathogenesis of alzheimer disease and other selected age-related neurodegenerative disorders.
        Free Radical Biol Med. 2017; 111: 253-261
        • Li R.
        • Zhang Y.
        • Rasool S.
        • et al.
        Effects and underlying mechanisms of bioactive compounds on type 2 diabetes mellitus and Alzheimer's disease.
        Oxid Med Cell Longev. 2019; 20198165707
        • Nita M.
        • Grzybowski A.
        The Role of the reactive oxygen species and oxidative stress in the pathomechanism of the age-related ocular diseases and other pathologies of the anterior and posterior eye segments in adults.
        Oxid Med Cell Longev. 2016; 20163164734
        • Lejri I.
        • Agapouda A.
        • Grimm A.
        • et al.
        Mitochondria- and oxidative stress-targeting substances in cognitive decline-related disorders: from molecular mechanisms to clinical evidence.
        Oxid Med Cell Longev. 2019; 20199695412
        • Yan M.H.
        • Wang X.
        • Zhu X.
        Mitochondrial defects and oxidative stress in Alzheimer disease and Parkinson disease.
        Free Radical Biol Med. 2013; 62: 90-101
        • Ahmad W.
        • Ijaz B.
        • Shabbiri K.
        • Ahmed F.
        • Rehman S.
        Oxidative toxicity in diabetes and Alzheimer's disease: mechanisms behind ROS/ RNS generation.
        J Biomed Sci. 2017; 24: 76
        • Bettcher B.M.
        • Kramer J.H.
        Inflammation and clinical presentation in neurodegenerative disease: a volatile relationship.
        Neurocase. 2013; 19: 182-200
        • Brown K.L.
        • Cosseau C.
        • Gardy J.L.
        • et al.
        Complexities of targeting innate immunity to treat infection.
        Trends Immunol. 2007; 28: 260-266
        • Bostanciklioglu M.
        An update on the interactions between Alzheimer's disease, autophagy and inflammation.
        Gene. 2019; 705: 157-166
        • McGeer E.G.
        • McGeer P.L.
        Neuroinflammation in Alzheimer's disease and mild cognitive impairment: a field in its infancy.
        J Alzheimers Dis. 2010; 19: 355-361
        • Ginhoux F.
        • Lim S.
        • Hoeffel G.
        • et al.
        Origin and differentiation of microglia.
        Front Cell Neurosci. 2013; 7: 45
        • Subhramanyam C.S.
        • Wang C.
        • Hu Q.
        • et al.
        Microglia-mediated neuroinflammation in neurodegenerative diseases.
        Semin Cell Dev Biol. 2019; 94: 112-120
        • Bajwa E.
        • Pointer C.B.
        • Klegeris A.
        The role of mitochondrial damage-associated molecular patterns in chronic neuroinflammation.
        Mediators Inflamm. 2019; 20194050796
        • Schwab C.
        • McGeer P.L.
        Inflammatory aspects of Alzheimer disease and other neurodegenerative disorders.
        J Alzheimers D. 2008; 13: 359-369
        • Mielke J.G.
        • Taghibiglou C.
        • Liu L.
        • et al.
        A biochemical and functional characterization of diet-induced brain insulin resistance.
        J Neurochem. 2005; 93: 1568-1578
        • Arnold S.E.
        • Arvanitakis Z.
        • Macauley-Rambach S.L.
        • et al.
        Brain insulin resistance in type 2 diabetes and Alzheimer disease: concepts and conundrums.
        Nat Rev Neurol. 2018; 14: 168-181
        • Desai G.S.
        • Zheng C.
        • Geetha T.
        • et al.
        The pancreas-brain axis: insight into disrupted mechanisms associating type 2 diabetes and Alzheimer's disease.
        J Alzheimers D. 2014; 42: 347-356
        • Rodriguez-Casado A.
        • Toledano-Diaz A.
        • Toledano A.
        Defective insulin signalling, mediated by inflammation, connects obesity to alzheimer disease; relevant pharmacological therapies and preventive dietary interventions.
        Curr Alzheimer Res. 2017; 14: 894-911
        • De Felice F.G.
        • Ferreira S.T.
        Inflammation, defective insulin signaling, and mitochondrial dysfunction as common molecular denominators connecting type 2 diabetes to Alzheimer disease.
        Diabetes. 2014; 63: 2262-2272
        • Phiel C.J.
        • Wilson C.A.
        • Lee V.M.
        • et al.
        GSK-3alpha regulates production of Alzheimer's disease amyloid-beta peptides.
        Nature. 2003; 423: 435-439
        • Tumminia A.
        • Vinciguerra F.
        • Parisi M.
        • et al.
        Type 2 diabetes mellitus and Alzheimer's disease: role of insulin signalling and therapeutic implications.
        Int J Mol Sci. 2018; 19
        • Li L.
        • Holscher C.
        Common pathological processes in Alzheimer disease and type 2 diabetes: a review.
        Brain Res Rev. 2007; 56: 384-402
        • Zhang Y.
        • Zhu H.
        • Mitchell J.R.
        • et al.
        T2 MRI texture analysis is a sensitive measure of tissue injury and recovery resulting from acute inflammatory lesions in multiple sclerosis.
        Neuroimage. 2009; 47: 107-111
        • de Carvalho Alegro M.
        • Valotta Silva A.
        • Yumi Bando S.
        • et al.
        Texture analysis of high resolution MRI allows discrimination between febrile and afebrile initial precipitating injury in mesial temporal sclerosis.
        Magn Reson Med. 2012; 68: 1647-1653
        • Alobaidli S.
        • McQuaid S.
        • South C.
        • et al.
        The role of texture analysis in imaging as an outcome predictor and potential tool in radiotherapy treatment planning.
        Br J Radiol. 2014; 8720140369
        • Chitalia R.D.
        • Kontos D.
        Role of texture analysis in breast MRI as a cancer biomarker: a review.
        J Magn Reson Imaging. 2019; 49: 927-938
        • Dragic M.
        • Zaric M.
        • Mitrovic N.
        • et al.
        Application of gray level co-occurrence matrix analysis as a new method for enzyme histochemistry quantification.
        Microsc microanal. 2019; 25: 690-698
        • Soni N.
        • Priya S.
        • Bathla G.
        Texture analysis in cerebral gliomas: a review of the literature.
        AJNR Am J Neuroradiol. 2019; 40: 928-934
        • Kassner A.
        • Thornhill R.E.
        Texture analysis: a review of neurologic MR imaging applications.
        AJNR Am J Neuroradiol. 2010; 31: 809-816
        • Sorensen L.
        • Shaker S.B.
        • de Bruijne M.
        Quantitative analysis of pulmonary emphysema using local binary patterns.
        IEEE Trans Med Imaging. 2010; 29: 559-569
        • Reig B.
        • Heacock L.
        • Geras K.J.
        • et al.
        Machine learning in breast MRI.
        J Magn Reson Imaging. 2019; https://doi.org/10.1002/jmri.26852
        • Ahmed M.R.
        • Zhang Y.
        • Feng Z.
        • et al.
        Neuroimaging and machine learning for dementia diagnosis: recent advancements and future prospects.
        IEEE Rev Biomed Eng. 2019; 12: 19-33
        • Handelman G.S.
        • Kok H.K.
        • Chandra R.V.
        • et al.
        eDoctor: machine learning and the future of medicine.
        J Intern Med. 2018; 284: 603-619
        • Simoes R.
        • van Cappellen van Walsum A.M.
        • Slump C.H.
        Classification and localization of early-stage Alzheimer's disease in magnetic resonance images using a patch-based classifier ensemble.
        Neuroradiology. 2014; 56: 709-721
        • Feng F.
        • Wang P.
        • Zhao K.
        • et al.
        Radiomic features of hippocampal subregions in Alzheimer's disease and amnestic mild cognitive impairment.
        Front Aging Neurosci. 2018; 10: 290
        • de Oliveira M.S.
        • Balthazar M.L.
        • D'Abreu A.
        • et al.
        MR imaging texture analysis of the corpus callosum and thalamus in amnestic mild cognitive impairment and mild Alzheimer disease.
        AJNR Am J Neuroradiol. 2011; 32: 60-66
        • Zhang J.
        • Yu C.
        • Jiang G.
        • et al.
        3D texture analysis on MRI images of Alzheimer's disease.
        Brain Imaging Behav. 2012; 6: 61-69
        • Gao N.
        • Tao L.X.
        • Huang J.
        • et al.
        Contourlet-based hippocampal magnetic resonance imaging texture features for multivariant classification and prediction of Alzheimer's disease.
        Metab Brain Dis. 2018; 33: 1899-1909
        • Luk C.C.
        • Ishaque A.
        • Khan M.
        • et al.
        Alzheimer's disease: 3-dimensional MRI texture for prediction of conversion from mild cognitive impairment.
        Alzheimers Dement. 2018; 10: 755-763
        • Oppedal K.
        • Engan K.
        • Eftestøl T.
        • et al.
        Classifying Alzheimer's disease, Lewy body dementia, and normal controls using 3D texture analysis in magnetic resonance images.
        Biomed Signal Process Control. 2017; 33: 19-29
        • Kovalev V.A.
        • Kruggel F.
        • Gertz H.J.
        • et al.
        Three-dimensional texture analysis of MRI brain datasets.
        IEEE Trans Med Imaging. 2001; 20: 424-433
        • Fletcher P.T.
        • Powell S.
        • Foster N.L.
        • et al.
        Quantifying metabolic asymmetry modulo structure in Alzheimer's disease.
        in: Information processing in medical imaging: proceedings of the... conference 20. 2007: 446-457
        • Heister D.
        • Brewer J.B.
        • Magda S.
        • et al.
        Predicting MCI outcome with clinically available MRI and CSF biomarkers.
        Neurology. 2011; 77: 1619-1628
        • Tabatabaei-Jafari H.
        • Shaw M.E.
        • Walsh E.
        • et al.
        Cognitive/functional measures predict Alzheimer's disease, dependent on hippocampal volume.
        J Gerontol B Psychol Sci Soc Sci. 2019; https://doi.org/10.1093/geronb/gbz011
        • Ferreira L.K.
        • Diniz B.S.
        • Forlenza O.V.
        • et al.
        Neurostructural predictors of Alzheimer's disease: a meta-analysis of VBM studies.
        Neurobiol Aging. 2011; 32: 1733-1741
        • Hansen D.
        • Ling H.
        • Lashley T.
        • et al.
        Review: clinical, neuropathological and genetic features of Lewy body dementias.
        Neuropathol Appl Neurobiol. 2019; 45: 335-654
        • Oppedal K.
        • Eftestol T.
        • Engan K.
        • et al.
        Classifying dementia using local binary patterns from different regions in magnetic resonance images.
        Int J Biomed Imaging. 2015; 2015572567
        • Kodama N.
        • Kawase Y.
        Computerized method for classification between dementia with Lewy bodies and Alzheimer's disease by use of texture analysis on brain MRI.
        in: World Congress on Medical Physics and Biomedical Engineering, Munich, Germany2009
        • Patel N.
        • Henry A.
        • Scarsbrook A.
        The value of MR textural analysis in prostate cancer.
        Clin Radiol. 2018; 74: 876-885
        • Leijenaar R.T.
        • Nalbantov G.
        • Carvalho S.
        • et al.
        The effect of SUV discretization in quantitative FDG-PET Radiomics: the need for standardized methodology in tumor texture analysis.
        Sci Rep. 2015; 5: 11075
        • Ford J.
        • Dogan N.
        • Young L.
        • Yang F.
        Quantitative radiomics: impact of pulse sequence parameter selection on MRI-based textural features of the brain.
        Contrast Media Mol Imaging. 2018; 20181729071
        • Waugh S.A.
        • Lerski R.A.
        • Bidaut L.
        • et al.
        The influence of field strength and different clinical breast MRI protocols on the outcome of texture analysis using foam phantoms.
        Med Phys. 2011; 38: 5058-5066
        • Solomon J.
        • Mileto A.
        • Nelson R.C.
        • et al.
        Quantitative features of liver lesions, lung nodules, and renal stones at multi-detector row CT examinations: dependency on radiation dose and reconstruction algorithm.
        Radiology. 2016; 279: 185-194
        • Suzuki K.
        Overview of deep learning in medical imaging.
        Radiol Phys Technol. 2017; 10: 257-273
        • Gillies R.J.
        • Kinahan P.E.
        • Hricak H.
        Radiomics: images are more than pictures, they are data.
        Radiology. 2016; 278: 563-577
        • Mathys H.
        • Davila-Velderrain J.
        • Peng Z.
        • et al.
        Single-cell transcriptomic analysis of Alzheimer's disease.
        Nature. 2019; 570: 332-337