Multiscale analysis enables comprehensive exploration of the hierarchical topology of brain networks. However, many of the existing brain network modeling methods average blood oxygenation level-dependent signals at the brain region level, thereby ignoring heterogeneity of voxels within each region of interest. Existing methods, such as Contrastive Voxel Clustering [1], have made progress in voxel-level analysis. However, they assume temporal stationarity, ignoring the spatial consistency nature of the brain, and require manual tuning of cluster numbers. To incorporate spatial consistency at the voxel level, we propose a semantically enhanced contrastive voxel clustering framework, which incorporates a spatial perspective based on regional homogeneity, aligning with principles of brain organization. To avoid manual cluster tuning, we introduce an adaptive voxel grouping module based on DeepDPM [2] for unsupervised clustering modification. Our proposed method outperforms baseline models on multiple datasets, validating effectiveness and generalization ability.

Amyotrophic Lateral Sclerosis (ALS) is a fatal neurodegenerative disorder characterized by rapid progression and poor prognosis. Early diagnosis remains highly challenging due to nonspecific clinical manifestations, leading to persistently high misdiagnosis rates. This study investigates the feasibility of using resting-state electromyography (EMG) signals for accurate ALS identification. Clinically collected EMG signals vary in length across individuals and muscle regions; further, preprocessing via sliding window methods yields inconsistent numbers of segmented samples, posing challenges for traditional neural networks to adapt to variable-length sequences. To address these issues, we propose a unified time-frequency learning framework called TF-Transformer. Centered on a Transformer-based feature extraction module, the framework treats segmented EMG data as sequence data, enabling efficient processing of variable-length sequences without padding or truncation. It synchronously extracts and fuses time and frequency domain features into a unified EMG representation for ALS identification. Experiments using five-fold cross-validation show that this method effectively utilizes the inherent time-frequency characteristics of resting state electromyographic signals, providing new technical insights and methods for clinical ALS diagnosis.

The Segment Anything Model (SAM) has demonstrated strong and versatile segmentation capabilities, along with intuitive prompt-based interactions. However, customizing SAM for medical image segmentation requires massive amounts of pixel-level annotations and precise point- or box-based prompt designs. To address these challenges, we introduce PGP-SAM, a novel prototype-based few-shot tuning approach that uses limited samples to replace tedious manual prompts. Our key idea is to leverage inter- and intra-class prototypes to capture class-specific knowledge and relationships. We propose two main components: (1) a plug-and-play contextual modulation module that integrates multi-scale information, and (2) a class-guided cross-attention mechanism that fuses prototypes and features for automatic prompt generation. Experiments on a public multi-organ dataset and a private ventricle dataset demonstrate that PGP-SAM achieves superior mean Dice scores compared with existing prompt-free SAM variants, while using only 10% of the 2D slices.

Analyzing functional connectivity patterns from resting-state functional magnetic resonance imaging (fMRI) requires unraveling its interrelations across spatial, temporal, and frequency domains. To comprehensively analyze four-dimensional (4D) fMRI data, we propose the Spatiotemporal Frequency Graph Neural Network (STFreqGNN), which processes dynamic heterogeneous graphs across spatial, temporal, and frequency domains using a transformer-style architecture. To reduce the complexity of multi-domain analysis with small sample sizes for fMRI datasets and ensure domainspecific interpretability, we introduce two structure-informed modules in the spatial and temporal domains to improve knowledge aggregation within each domain. Specifically, the plugin GNNs transmit information within the static homogeneous brain region graphs, and recurrent blocks aggregate features from heterogeneous nodes defined across different temporal windows. Additionally, we design cross-domain masked self-attention blocks to prevent attention captured by irrelevant or redundant token pairs, finally enabling efficient disease-specific feature learning. Experimental results on both public and in-house datasets suggest that the proposed method is not only superior to several state-of-the-art methods on fMRI-based classification but also preserves interpretation ability in all these domains.

Background and objectiveWe investigated the association ofAPOEalleles with CT-based cerebral amyloid angiopathy (CAA) markers including subarachnoid extension (SAE) and finger-like projection (FLP).MethodsWe included patients with acute primary supratentorial intracerebral haemorrhage (ICH) from a multicentre cohort in China. First, the association ofAPOEwith ICH location (lobar vs non-lobar) was evaluated. Next, the relationships ofAPOEwith SAE, FLP, and the coexistence of the two (SAE+FLP) were evaluated.Results533 patients with supratentorial ICH were enrolled. Among them were 138 patients with lobar ICH and 395 with non-lobar ICH. Compared with the non-lobar group,APOEε4 (OR 1.894, 95% CI 1.138 to 3.154, p=0.014) and ε2/ε4 (OR 6.098, 95% CI 1.414 to 26.293, p=0.015) were associated with lobar ICH. With regard to CAA markers,APOEε2 was associated with SAE (OR 2.109, 95% CI 1.167 to 3.810, p=0.013), ε4 was associated with FLP and SAE+FLP (OR 3.026, 95% CI 1.353 to 6.767, p=0.007; OR 3.514, 95% CI 1.485 to 8.316, p=0.004, respectively) and ε2/ε4 was associated with all three factors (SAH: OR 7.599, 95% CI 1.764 to 32.734, p=0.006; FLP: OR 20.333, 95% CI 3.278 to 126.137, p=0.001; SAE+FLP: OR 30.568, 95% CI 4.460 to 209.503, p<0.001) after adjusting for age, and remained significant after adjusting for age and ICH volume.ConclusionIn patients with spontaneous supratentorial ICH,APOEε2 and ε4 alleles were associated with SAE and FLP, respectively, suggestingAPOEallele-specific effects on CT markers of CAA and their potential mechanisms.

Tracking a patient's cognitive status early in the onset of the disease provides an opportunity to diagnose and intervene in Alzheimer's disease (AD). However, relying solely on magnetic resonance imaging (MRI) images with traditional classification and regression models may not fully extract finer-grained information. This study proposes a multi-task Fusion Language Image Question Answering model (FLIQA-AD) to perform AD identification and Mini Mental State Examination (MMSE) prediction. Specifically, a 3D Adapter is introduced in Vision Transformer (ViT) model for image feature extraction. The patient electronic health records (EHR) information and questions related to the disease work as text prompts to be encoded. Then, an AD-Former model, which combines self-attention and cross-attention mechanisms, is used to capture the correlation between EHR information and structure features. After that, the extracted brain structural information and textual content are combined as input sequences for the large language model (LLM) to identify AD and predict the corresponding MMSE score. Experimental results demonstrate the strong discrimination and MMSE prediction performance of the model, as well as question-answer capabilities.

BackgroundThe island sign is a predictor of hematoma expansion and worse outcomes in patients of spontaneous primary intracerebral hemorrhage (ICH). The biological mechanism of the island sign remains unclear, but its presence might be influenced by the underlying vasculopathy related to Apolipoprotein E (APOE) genotypes. Therefore, we aimed to research the association between APOE genotypes and the island sign.MethodsWe enrolled patients with primary supratentorial ICH in a multicenter cohort in northern China with baseline noncontrast CT images performed within 14 days after symptoms onset and APOE genotype available. The island sign was rated on the CT images according to validated criteria. Univariable and multivariable analyses were used to identify the association between APOE genotypes and the island sign, stratified by the ICH location.ResultsAmong 460 patients enrolled, 122 were lobar ICH. In all patients, after adjusting for age, sex, hypertension, and time to CT, the presence of the APOE ε4 allele (OR 2.020, 95% CI 1.064–3.834, p = 0.032) was associated with the island sign, whereas the presence of the APOE ε2 allele (OR 0.734, 95% CI 0.339–1.593, p = 0.435) was not. After stratifying by ICH location, multivariable analysis revealed that APOE ε4 (OR 3.510, 95% CI 1.393–8.846, p = 0.008), rather than ε2 (OR 0.621, 95% CI 0.203–1.901, p = 0.404), was associated with the island sign in lobar ICH patients. Neither the ε2 nor the ε4 allele was associated with the island sign among nonlobar ICH patients.ConclusionThe APOE ε4 allele was associated with the island sign in lobar ICH patients. Our findings indicate that the presence of the island sign may be influenced by the underlying vasculopathy related to APOE ε4, which increases amyloid deposition in the cerebral vasculature.

Lobar intracerebral hemorrhage (ICH) is associated with a high risk of recurrence, particularly in elderly patients, where cerebral amyloid angiopathy (CAA) is often the primary cause. Diagnostic markers of CAA-related ICH, including subarachnoid hemorrhage (SAH) and finger-like projection (FLP), have recently been developed. Here, we aimed to explore the associations between SAH, FLP and the risk of ICH recurrence in lobar ICH patients. We analyzed data from consecutive lobar ICH patients using the method of cohort study. We divided them into 4 groups on the basis of the presence or absence of SAH and FLP on CT imaging. The Cox regression model and competing risk model were used to analyze the associations of SAH and FLP with the risk of ICH recurrence at 1 year. In total, 353 patients with lobar ICH (median age 74 [62, 81] years, 57.2

The structural Magnetic resonance imaging (sMRI) of spinal cord plays a significant role in the clinical diagnosis of Amyotrophic Lateral Sclerosis (ALS). But due to small cross-sectional area in the axial plane and long sagittal/coronal expansion of spinal cord, the diagnosis of ALS using sMRI of spinal cord has remained largely at the stage of morphological observation. In this study, a Multi-branch attention-based deep learning method is proposed to solve this problem. Multi-branch framework is utilized to extract general features of all levels of spinal cord for challenging of long sagittal and coronal expansion of spinal cord, and attention module coupled with multi-scale module in each branch is applied to extract multi-scale features and pay more attention to the important regions of the spinal cord in the axial plane. Experiments show that the proposed method obtains better performance in ALS identification, which implies that the proposed method can extract features of important region in the spinal cord and could be helpful to find more regions sensitive for ALS disease identification.

Resting-state functional magnetic resonance imaging (fMRI) provides an efficient way to analyze the functional connectivity between brain regions. A comprehensive understanding of brain functionality requires a unified description of multi-scale layers of neural structure. However, existing brain network modeling methods often simplify this property by averaging Blood oxygen level dependent (BOLD) signals at the brain region level for fMRI-based analysis with the assumption that BOLD signals are homogeneous within each brain region, which ignores the heterogeneity of voxels within each Region of Interest (ROI). This study introduces a novel multi-stage self-supervised learning framework for multiscale brain network analysis, which effectively delineates brain functionality from voxel to ROIs and up to sample level. A Contrastive Voxel Clustering (CVC) module is proposed to simultaneously learn the voxel-level features and clustering assignments, which ensures the retention of informative clustering features at the finest voxel-level and concurrently preserves functional connectivity characteristics. Additionally, based on the extracted features and clustering assignments at the voxel level by CVC, a Brain ROI-based Graph Neural Network (BR-GNN) is built to extract functional connectivity features at the brain ROI-level and used for sample-level prediction, which integrates the functional clustering maps with the pre-established structural ROI maps and creates a more comprehensive and effective analytical tool. Experiments are performed on two datasets, which illustrate the effectiveness and generalization ability of the proposed method by analyzing voxel-level clustering results and brain ROIs-level functional characteristics. The proposed method provides a multiscale modeling framework for brain functional connectivity analysis, which will be further used for other brain disease identification. Code is available at https://github.com/yanliugroup/fmri-cvc.

Stroke and dementia have been linked to the appearance of white matter hyperintensities (WMHs). Meanwhile, diffusion tensor imaging (DTI) might capture the microstructural change in white matter early. Specific dietary interventions may help to reduce the risk of WMHs. However, research on the relationship between specific nutrients and white matter changes is still lacking. We aimed to investigate the causal effects of essential nutrients (amino acids, fatty acids, mineral elements, and vitamins) on WMHs and DTI measures, including fraction anisotropy (FA) and mean diffusivity (MD), by a Mendelian randomization analysis. We selected single nucleotide polymorphisms (SNPs) associated with each nutrient as instrumental variables to assess the causal effects of nutrient-related exposures on WMHs, FA, and MD. The outcome was from a recently published large-scale European Genome Wide Association Studies pooled dataset, including WMHs (N = 18,381), FA (N = 17,663), and MD (N = 17,467) data. We used the inverse variance weighting (IVW) method as the primary method, and sensitivity analyses were conducted using the simple median, weighted median, and MR-Egger methods. Genetically predicted serum calcium level was positively associated with WMHs risk, with an 8.1% increase in WMHs risk per standard deviation unit increase in calcium concentration (OR = 1.081, 95% CI = 1.006-1.161, p = 0.035). The plasma linoleic acid level was negatively associated with FA (OR = 0.776, 95% CI = 0.616-0.978, p = 0.032). Our study demonstrated that genetically predicted calcium was a potential risk factor for WMHs, and linoleic acid may be negatively associated with FA, providing evidence for interventions from the perspective of gene-environment interactions.

Background Computed tomography angiography (CTA) and magnetic resonance angiography (MRA) provide accurate vascular imaging information, but their use may be contraindicated. Color Doppler ultrasonography (CDU) provides simple, safe, noninvasive, and reproducible imaging. We therefore investigated the role of preoperative CDU combined with CTA and MRA in the quantification, typing, and diagnosis of carotid body tumors (CBTs). Methods We retrospectively analyzed patients with CBTs categorized into group A (type I [n = 1] and type II [n = 10]) or group B (type III [n = 56]) per the intraoperative Shamblin classification. CDU, CTA, and MRA characteristics of CBTs were observed, surgical results were correlated, and the diagnostic threshold of the CBT classification was calculated. Results CBTs were usually located at the common carotid artery bifurcation, encircling the carotid artery. An increased angle was found between the internal and external carotid arteries. On CDU, CBTs primarily presented as homogeneous hypoechoic masses with clear boundaries, rich flow signals, and a high-speed, low-resistance artery-like flow spectrum. CTA showed uniform or heterogeneous marked enhancement. MRA showed mixed T1 and slightly longer T2 signals and uniform or uneven obvious enhancement. With increases in the lesion size, amount of blood transfused, and operation time, the intraoperative classification level and possibility of skull-base invasion increased. When the maximum diameter of the lesion, the volume of the tumor, the distance between the upper margin of the tumor to the mastoid and the mandibular angle were 3.10 cm, 10.15 cm 3 , − 3.26 cm, and 0.57 cm, respectively, the largest Youden index was the best diagnostic boundary value for Shamblin type III tumors. Conclusions CDU combined with CTA and MRA can accurately evaluate the size and classification of CBTs.

BACKGROUND:Whether encoding or retrieval failure contributes to memory binding deficit in amnestic mild cognitive impairment (aMCI) has not been elucidated. Also, the potential brain structural substrates of memory binding remained undiscovered.OBJECTIVE:To investigate the characteristics and brain atrophy pattern of encoding and retrieval performance during memory binding in aMCI.METHODS:Forty-three individuals with aMCI and 37 cognitively normal controls were recruited. The Memory Binding Test (MBT) was used to measure memory binding performance. The immediate and delayed memory binding indices were computed by using the free and cued paired recall scores. Partial correlation analysis was performed to map the relationship between regional gray matter volume and memory binding performance.RESULTS:The memory binding performance in the learning and retrieval phases was worse in the aMCI group than in the control group (F = 22.33 to 52.16, all p < 0.001). The immediate and delayed memory binding index in the aMCI group was lower than that in the control group (p < 0.05). The gray matter volume of the left inferior temporal gyrus was positively correlated with memory binding test scores (r = 0.49 to 0.61, p < 0.05) as well as the immediate (r = 0.39, p < 0.05) and delayed memory binding index (r = 0.42, p < 0.05) in the aMCI group.CONCLUSION:aMCI may be primarily characterized by a deficit in encoding phase during the controlled learning process. Volumetric losses in the left inferior temporal gyrus may contribute to encoding failure.

Background and Objective: Both Alzheimer’s disease (AD) and Parkinson’s disease (PD) are progressive neurodegenerative diseases. Early identification is very important for the prevention and intervention of their progress. Hippocampus plays a crucial role in cognition, in which there are correlations between atrophy of Hippocampal subfields and cognitive impairment in neurodegenerative diseases. Exploring biomarkers in the prediction of early cognitive impairment in AD and PD is significant for understanding the progress of neurodegenerative diseases. Methods: A multi-scale attention-based deep learning method is proposed to perform computer-aided diagnosis for neurodegenerative disease based on Hippocampal subfields. First, the two dimensional (2D) Hippocampal Mapping Image (HMI) is constructed and used as input of three branches of the following network. Second, the multi-scale module and attention module are integrated into the 2D residual network to improve the diversity of the extracted features and capture significance of various voxels for classification. Finally, the role of Hippocampal subfields in the progression of different neurodegenerative diseases is analyzed using the proposed method. Results: Classification experiments between normal control (NC), mild cognitive impairment (MCI), AD, PD with normal cognition (PD-NC) and PD with mild cognitive impairment (PD-MCI) are carried out using the proposed method. Experimental results show that subfields subiculum, presubiculum, CA1, and molecular layer are strongly correlated with cognitive impairment in AD and MCI, subfields GC-DG and fimbria are sensitive in detecting early stage of cognitive impairment in MCI, subfields CA3, CA4, GC-DG, and CA1 show significant atrophy in PD. For exploring the role of Hippocampal subfields in PD cognitive impairment, we find that left parasubiculum, left HATA and left presubiculum could be important biomarkers for predicting conversion from PD-NC to PD-MCI. Conclusion: The proposed multi-scale attention-based network can effectively discover the correlation between subfields and neurodegenerative diseases. Experimental results are consistent with previous clinical studies, which will be useful for further exploring the role of Hippocampal subfields in neurodegenerative disease progression.

Background:The classification of calcaneofibular ligament (CFL) injuries on magnetic resonance imaging (MRI) is time-consuming and subject to substantial interreader variability. This study explores the feasibility of classifying CFL injuries using deep learning methods by comparing them with the classifications of musculoskeletal (MSK) radiologists and further examines image cropping screening and calibration methods.Methods:The imaging data of 1,074 patients who underwent ankle arthroscopy and MRI examinations in our hospital were retrospectively analyzed. According to the arthroscopic findings, patients were divided into normal (class 0, n=475); degeneration, strain, and partial tear (class 1, n=217); and complete tear (class 2, n=382) groups. All patients were divided into training, validation, and test sets at a ratio of 8:1:1. After preprocessing, the images were cropped using Mask region-based convolutional neural network (R-CNN), followed by the application of an attention algorithm for image screening and calibration and the implementation of LeNet-5 for CFL injury classification. The diagnostic effects of the axial, coronal, and combined models were compared, and the best method was selected for outgroup validation. The diagnostic results of the models in the intragroup and outgroup test sets were compared with those results of 4 MSK radiologists of different seniorities.Results:The mean average precision (mAP) of the Mask R-CNN using the attention algorithm for the left and right image cropping of axial and coronal sequences was 0.90-0.96. The accuracy of LeNet-5 for classifying classes 0-2 was 0.92, 0.93, and 0.92, respectively, for the axial sequences and 0.89, 0.92, and 0.90, respectively, for the coronal sequences. After sequence combination, the classification accuracy for classes 0-2 was 0.95, 0.97, and 0.96, respectively. The mean accuracies of the 4 MSK radiologists in classifying the intragroup test set as classes 0-2 were 0.94, 0.91, 0.86, and 0.85, all of which were significantly different from the model. The mean accuracies of the MSK radiologists in classifying the outgroup test set as classes 0-2 were 0.92, 0.91, 0.87, and 0.85, with the 2 senior MSK radiologists demonstrating similar diagnostic performance to the model and the junior MSK radiologists demonstrating worse accuracy.Conclusions:Deep learning can be used to classify CFL injuries at similar levels to those of MSK radiologists. Adding an attention algorithm after cropping is helpful for accurately cropping CFL images.

With the progression of neurodegenerative disease, functional connectivity between brain regions has changed, which can be reflected locally by Blood-oxygen-level-dependent (BOLD) signal measured in functional magnetic resonance imaging (fMRI). Most studies assume BOLD signals are homogeneous within brain regions, ignoring voxel-level changes. In this paper, we propose a novel framework for voxel-based feature extraction and recollection to characterize the BOLD signal and analyze the functional connectivity of brain networks and uncover biomarkers for abnormalities. Specifically, a weakly-supervised learning strategy is adopted to extract discriminative representation from original BOLD signals. Considering the heterogeneity of BOLD signals within brain regions of interest (ROIs), we employ an unsupervised-based deep clustering method to automatically recollect features to different clusters. Experiments on Alzheimer’s Disease (AD) recognition using Graph neural network (GNN) validate the effectiveness of our framework. To the best of our knowledge, this is the first work to consider BOLD signal heterogeneity for feature extraction to measure functional connectivity in GNN, which provides a voxel-level scenario that can be migrated to other tasks.

Background and Objective: For early identification of Alzheimer's disease (AD) based on multi-modal magnetic resonance imaging (MRI) data, it is important to make comprehensive use of image features and non-image information to analyze the gray matter atrophy and the structural/functional connectivity abnormalities for different courses of AD. Methods: In this study, we propose an extensible hierarchical graph convolutional network (EH-GCN) for early AD identification. Based on the extracted image features from multi-modal MRI data using the presented multi-branch residual network (ResNet), the brain regions-of-interests (ROIs) based GCN is built to extract structural and functional connectivity features between different ROIs of the brain. In order to further improve the performance of AD identification, an optimized spatial GCN is proposed as convolution operator in the population-based GCN to avoid rebuilding the graph network and take advantage of relationships between subjects. Finally, the proposed EH-GCN is built by embedding the image features and internal brain connectivity features into the spatial population-based GCN, which provides an extensible way to improve early AD identification performance by adding imaging features and non-image information from multi-modal data. Results: Experiments are performed on two datasets, which illustrate the effectiveness of the extracted structural/functional connectivity features and the high computational efficiency of the proposed method. The classification accuracy of AD vs NC, AD vs MCI and MCI vs NC classification tasks reaches 88.71%, 82.71% and 79.68% respectively. The extracted connectivity features between ROIs indicate that functional abnormalities are earlier than gray matter atrophy and abnormalities of structural connections, which is consistent with the clinical manifestations. The proposed method allows for the addition of other modal image features and non-image information from multi-modal data to continuously improve the performance of clinical data analysis. Conclusions: The proposed method can help us comprehensively analyze the role of gray matter atrophy, the damage of white matter nerve fiber tracts and the degradation of functional connectivity for different courses of AD, which could be useful for further extraction of clinical biomarkers for early AD identification.

Background and PurposeConvexity subarachnoid hemorrhage (cSAH) may predict an increased recurrence risk in cerebral amyloid angiopathy (CAA)-related intracerebral hemorrhage (ICH) survivors. We aimed to investigate whether cSAH detected on CT was related to early recurrence in patients with ICH related to CAA.MethodsWe analyzed data from consecutive lobar ICH patients diagnosed as probable or possible CAA according to the Boston criteria using the method of cohort study. Demographic and clinical data, ICH recurrence at discharge and within 90 days were collected. The association between cSAH detected on CT and early recurrent ICH was analyzed using multivariable logistic regression.ResultsA total of 197 cases (74 [66–80] years) were included. cSAH was observed on the baseline CT of 91 patients (46.2%). A total of 5.1% (10/197) and 9.5% (17/179) of patients experienced ICH recurrence within 2 weeks and 90 days, respectively. The presence of cSAH was related to recurrence within 2 weeks (OR = 5.705, 95%CI 1.070–30.412, P = 0.041) after adjusting for hypertension, previous symptomatic ICH and anticoagulant use. The presence of cSAH was related to recurrence within 90 days (OR 5.473, 95%CI 1.425–21.028, P = 0.013) after adjusting for hypertension, previous symptomatic ICH and intraventricular hemorrhage. The similar results were obtained in other models using different methods to select adjusting variables.ConclusionIn patients with lobar ICH related to CAA, 5.1% and 9.5% of them experienced ICH recurrence within 2 weeks and 90 days, respectively. CT-visible cSAH was detected in 46.2% of patients and indicates an increased risk for early recurrent ICH.

Hippocampus is an important anatomical region for Alzheimer's Disease (AD) identification. In this paper, a multi-scale attention-based convolutional network is proposed for AD identification. The two dimensional (2D) images in three different planes of hippocampal subfields are used as input of three branches of the proposed network, which achieves effective extraction of three dimensional (3D) data features while reducing the network complexity and improving the computational efficiency. The end-to-end 2D multi-scale attention-based deep learning network improves the diversity of the extracted features and captures significance of various voxels for classification, which achieves significant classification performance without handcrafted feature extraction and model stacking. Experimental results illustrate the effectiveness of the proposed method on AD identification. The proposed method will be useful for further medical analysis on hippocampal subfields of the brain for diagnosis of neurodegenerative disease.

Background and Objective: Alzheimer's Disease (AD) is a progressive irreversible neurodegeneration disease and thus timely identification is critical to delay its progression. Methods: In this work, we focus on the traditional branch to design discriminative feature extraction and selection strategies to achieve explainable AD identification. Specifically, a spatial pyramid based three-dimensional histogram of oriented gradient (3D-HOG) feature learning method is proposed. Both global and local texture changes are included in spatial pyramid 3D-HOG (SPHOG) features for comprehensive analysis. Then a modified wrapper-based feature selection algorithm is introduced to select the discriminative features for AD identification while reduce feature dimensions. Results: Discriminative SPHOG histograms with various resolutions are selected, which can represent the atrophy characteristics of cerebral cortex with promising performance. As subareas corresponding to selected histograms are consistent with clinical experience, explanatory is emphasized and illustrated with Hippocampus. Conclusion: Experimental results illustrate the effectiveness of the proposed method on feature learning based on samples obtained from common dataset and a clinical dataset. The proposed method will be useful for further medical analysis as its explanatory on other region-of-interests (ROIs) of the brain for early diagnosis of AD. (C) 2021 Elsevier B.V. All rights reserved.