Comparing Deep Learning CNN method with Traditional MRI-based Hippocampal segmentation and volumetry for Early Alzheimer’s Disease Diagnosis Across Diverse Populations

Comparing Deep Learning CNN method with Traditional MRI-based Hippocampal segmentation and volumetry for Early Alzheimer’s Disease Diagnosis Across Diverse Populations | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Comparing Deep Learning CNN method with Traditional MRI-based Hippocampal segmentation and volumetry for Early Alzheimer’s Disease Diagnosis Across Diverse Populations Nur Shahidatul Nabila Ibrahim, Subapriya Suppiah, Buhari Ibrahim, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6790322/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Dec, 2025 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract The advent of artificial intelligence (AI) driven software has impacted numerous aspects of medicine, leading to automated algorithms that assist in performing feature extraction, making measurements on diagnostic imaging, and aiding in diagnosing disorders. AI-based convoluted neural networks (CNN) enable automated segmentation of the hippocampal volume seen on MRI diagnostic imaging, hence facilitating the diagnosis of Alzheimer’s disease (AD). Traditional voxel-based morphometry (VBM) used for measuring hippocampal volume can be time-laborious, thus CNN-based algorithms can minimize the time and reduce human errors. We utilized HippoDeep, an open-source CNN-based algorithm, to compare the hippocampal datasets from a Caucasian population with a dataset from a Southeast Asian AD and cognitively healthy control (HC) population. ROC analysis revealed superior diagnostic performance for HippoDeep, with AUCs of 0.918 (left hippocampus) and 0.882 (right hippocampus), compared to VBM’s 0.788 and 0.741, respectively. We determined cut-off thresholds for hippocampal volume to further improve the classification method. CNN-based method outperformed traditional semiautomated for segmentation accuracy (p < 0.001) with insignificant interpopulation differences. Moreover, CNN-derived hippocampal volumes exhibited stronger correlations with MMSE scores (r = 0.63 vs. r = 0.42). HippoDeep offers accurate, reproducible, and generalizable hippocampal segmentation, supporting its potential as a clinical tool for early AD diagnosis across diverse populations. Biological sciences/Neuroscience Biological sciences/Neuroscience/Cognitive ageing Health sciences/Diseases/Neurological disorders/Dementia Figures Figure 1 Figure 2 Figure 3 Introduction Alzheimer’s disease (AD) and other dementias have progressively become more prevalent in Southern Asia due to the growing elderly population, which has placed a greater burden on the healthcare systems (Zhu et al., 2023). Research reveals that over 3.55 million individuals in Southeast Asia are afflicted with dementia, accounting for 8.5% of older adults in Malaysia living with AD, i.e. approximately 260,000 people and projected to rise to 590,000 by 2050 (Alzheimer’s Disease International, 2020). According to Alzheimer's Disease International (ADI), the number of individuals with dementia in Indonesia is expected to reach nearly 4 million by 2050 (Leow et al., 2024 ). The need to monitor the trends in the prevalence of dementia as populations age with increasing life expectancy is paramount for health care planning and resource allocation in Singapore (Subramaniam et al., 2025 ). However, there has been no standard monitoring of this population. Moreover, insufficient awareness of AD has caused it to become the subject of stigma and misunderstanding among the general public, which may impede early detection and therapy. AD is a neurodegenerative condition that is characterised by progressive cognitive decline and distinctive symptoms like brain shrinkage and beta-amyloid accumulation (Raji et al., 2024 ). Hippocampal atrophy is a hallmark biomarker of AD and other forms of dementia, as well as cerebrovascular accidents (CVA), underscoring its clinical relevance in neurodegenerative and vascular pathologies (Gemmell et al., 2011 ; Vejandla et al., 2024 ). Conversely, a correlation was discovered between PET-Aβ and overall hippocampal atrophy in cognitively intact elderly persons (Xia et al., 2024). Aβ has a more pronounced impact on hippocampus degeneration in the initial phases of the disease (Rao et al., 2024 ), whereas other conditions contribute more significantly at the later stages of dementia. Particularly in normal aging, the hippocampal volume decreases at a gradual rate, approximately 1–2% per year after the age of 60. This mild shrinkage is typically associated with age-related cognitive decline but does not significantly impair daily functioning. In contrast, AD is characterized by accelerated hippocampal atrophy, with rates ranging from 3–5% per year in the early stages and potentially higher as the disease progresses. This rapid loss is strongly correlated with declining episodic memory and cognitive performance, distinguishing AD from normal aging (Sghirripa et al., 2025). Traditionally, structural magnetic resonance imaging (sMRI) is the preferred non-invasive detection tool to identify the features of AD during the the later stages, however, this can be challenging in the early phase, despite a priori knowledge of hippocampal atrophy and cortical thinning being associated with cognitive decline in AD (Mostafa et al., 2023 ). T1 MPRAGE MRI scans, in the coronal orthogonal plane, provide comprehensive visualisation of structural alterations in the brain, enabling precise evaluation and monitoring of disease progression (Lavielle et al., 2022 ). This imaging technique is crucial for the early detection and observation of changes in the brain related to AD. Quantitative sMRI can forecast the progression to dementia in individuals with moderate cognitive impairment. Moreover, quantitative sMRI can yield critical insights for prediction of disease and also help dictate strategies for early interventions. Integrating this imaging technology into clinical practice can enhance diagnostic precision and patient outcomes for those who are predisposed to dementia (Loftus et al., 2023 ). Given the strong association between hippocampal volume reduction and cognitive impairment, accurate segmentation of the hippocampus using sMRI is crucial for early diagnosis, monitoring, and research applications. While manual segmentation remains the gold standard due to its anatomical precision, it is time-consuming, operator-dependent, and often lacks reproducibility across raters Voxel-based morphometry (VBM) is a semi-automated neuroimaging methodology employed to evaluate structural variations in the brain, with has encompassed alterations in the hippocampus volume. Although VBM is proficient for group-level analyses, it is less precise at the individual level in comparison to automated segmentation techniques. Fully automated segmentation methods using deep learning (DL) algorithms such as convolutional neural networks (CNN), have led to the development of various softwares. Multiple automated hippocampal segmentation methods have emerged over the past two decades. These include atlas-based techniques such as FreeSurfer, which have been extensively validated but show limitations in populations with brain atrophy or lesions, such as those following CVA. A significant drawback of many of these algorithms is their reliance on young, healthy brain templates, lack of generalizability to aging and diseased brains, and operational demands in terms of computational resources and user expertise. Initial studies(Thyreau et al., 2018 ) have shown that HippoDeep, which was developed using FreeSurfer, has been used successfully in non-demented (non-AD) subjects, providing reliable hippocampal volume assessments. The efficacy in bigger AD populations has yet to be comprehensively evaluated. Nevertheless, HippoDeep's capacity to efficiently process extensive datasets renders it advantageous for wider applicability in AD research. Additional validation in Alzheimer's disease-specific cohorts is essential to confirm its diagnostic utility. While HippoDeep has been successfully employed for automated hippocampal segmentation and measurement using normative dataset of stroke and AD patients from the Western populations (Sghirripa et al., 2025), its performance among AD subjects in an Asian population dataset has not been comprehensively assessed. This may be due to a lack of information regarding the availability of imaging technologies, an inadequacy of training skills for interpretation, and the requirement of high computational resources for the evaluations of neurodegenerative disorders (Schöll et al., 2024 ). Hence, HippoDeep is recommended as the preferred CNN-based automated hippocampal segmentation and volumetry method that’s helps to reduce the interpretation time and enhance the standardization of brain measurements for better longitudinal comparison. Despite their theoretical advantages, CNN-based segmentation methods such as HippoDeep have yet to be widely validated in non-Western populations and under clinical conditions common in resource-constrained settings. We aimed to determine the difference in the range of hippocampal volume in a Southeast Asian AD population dataset, comprising of people from an Austronesian group indigenous to the Malay Peninsula, compared with dataset from a Caucasian population from the ADNI database. We also evaluated the diagnostic accuracy of the CNN-based HippoDeep segmentation model with traditional VBM method in assessing hippocampal volume among AD patients and healthy controls (HC) in a Caucasian compared to a Southeast Asian population, by establishing hippocampal volume cut-off thresholds, using ROC curve analysis. Subsequently, we assessed the generalizability and diagnostic performance of HippoDeep in a diverse population. Finally, we explored the relationship between hippocampal volume loss and cognitive impairment, as measured by Mini-Mental State Examination (MMSE) scores, to validate the clinical relevance of this CNN-based automated segmentation tool. Methods Ethics information This study was approved by the Malaysian national research ethics committee, which is the Medical Research Ethics Committee (MREC) of National Medical Registration Registry (NMRR) Malaysia (NMRR-19-2719-49105) and the Ethics Committee for Research Involving Human Subjects of Universiti Putra Malaysia (JKEUPM-2019-328). All the subjects from this nationally conducted research gave their informed consent, which was obtained from the participants and/or their legal guardians. Furthermore, the research was performed in accordance with the Declaration of Helsinki. Design This study utilized a dataset comprising of Caucasian subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu) and also data of Southeast Asian subjects from a national Fundamental Research Grant Scheme (FRGS) project in Malaysia. Participants diagnosed with AD were identified based on clinical criteria and underwent multiple neuroimaging evaluations, allowing for a comprehensive analysis of disease progression and biomarker changes over time. Neuropsychological assessment included a range of cognitive evaluation tools, such as the Mini-Mental State Examination (MMSE). Cognitively healthy control (HC) subjects in the Southeast Asian study were recruited through advertisements on community bulletin boards and distributed fliers. The dataset for Southeast Asian subjects is in the process of being deposited in our academic institutional repository. We evaluated the performance of hippocampal segmentation using both deep learning (DL)-based and traditional atlas-based methods on sMRI scans from two demographically distinct datasets: the Alzheimer's Disease Neuroimaging Initiative (ADNI) and a Southeast Asian FRGS study memory clinic cohort. CNN-based segmentation was conducted using HippoDeep, while VBM method using SPM Mathlab represented the traditional atlas-based approach. The ADNI database solely provides gender and age demographic data. Age is continuous, while gender is female or male. The findings may not apply to larger populations due to this constraint. For analysis, data pre-processing standardised datasets from multiple sources and imaging data. Pre-processing neuroimaging data corrected head motion and standardised picture intensity across sessions. The Dice Similarity Coefficient (DSC) was used to compare automated and manual segmentations on 30 representative images from each cohort to assess segmentation quality. Statistical analysis includes ANOVA for group comparisons and Pearson correlation for hippocampus size and MMSE scores. This study compared bilateral hippocampus volume measurements from VBM and HippoDeep. Additionally, we examined the link between cognitive performance evaluations and hippocampal volume. HippoDeep and VBM were evaluated for AD diagnosis accuracy using ROC curve with the MMSE scores combined with clinical assessment criteria as a reference standard. Sampling plan The methodology employed for hippocampal volume segmentation in this study was derived from prior research (Thyreau et al., 2018 ). The authors relied on the availability of FreeSurfer (RRID:SCR_001847) software and pre-existing FreeSurfer-labeled online datasets as a crucial knowledge store. This facilitates the swift and significant enhancement of training examples (Korbmacher et al., 2024 ). Transfer learning is a machine learning approach that involves training a model using the results of a sophisticated algorithm (Aboutorab et al., 2021 ). In deep learning, transfer learning typically refers to the fine-tuning of a model that has been pre-trained on a larger, more diverse dataset, which is then utilised in a more specialised area with restricted datasets (Zhao et al., 2024 ). The researchers utilised simulated pictures of arterial trees produced by a physiologically accurate model to get a high level of precision and robustness (Bolchini et al., 2023 ; Xing et al., 2024 ). The simulated photos functioned as the input for the classifier's training and evaluation. The installation of HippoDeep Brain hippocampus segmentation was conducted using a repository compatible with Ubuntu/Linux and Mac OS: https://github.com/bthyreau/hippodeep_pytorch . We employed the executable Windows pre-compiled application, which is a plug-and-play application. The pre-compiled application designed for the Windows platform can be accessed at: https://drive.usercontent.google.com/download?id=1rSp0S7kCR8ksPiFuujbhbDe0L-qHbmdx&export=download&authuser=0 (Ibrahim et al., 2023 ). The VBM procedure(Huang et al., 2023 ) initiated with segmentation (spatial pre-processing): employing the native space option, a tissue class image aligned with the original was generated, subsequently followed by the implementation of Dartel(Sui et al., 2023 ) to produce templates from selected images for distortion correction. The next process was normalisation, during which the tissue class image was transformed into a standardised space, followed by smoothing to improve the signal-to-noise ratio. These techniques are crucial for data preparation before subsequent analysis. Imagesgit with similar dimensions, orientation, and voxel size were selected for statistical analysis using a factorial design specification. The value was calculated via a design specification file from the computer's file system. The Automated Anatomical Labelling (AAL) toolkit(Khagi et al., 2021 ) was superimposed onto the Montreal Neurological Institute (MNI) template(Gaser et al., 2024 ) to derive the volume at the cluster level. Analysis Plan All the T1-weighted images were preprocessed using SPM Toolbox ( https://www.fil.ion.ucl.ac.uk/spm/download/restricted/eldorado/spm12.zip ) implemented in MATLAB (R2021a) (Di & Biswal, 2023 ; Tavares et al., 2020 ). Data availability We are committed to share our raw data and materials on acceptance of our Stage 2 manuscript. Our data is deposited in a self-repository (pending complete institutional repository upload) and is available to be viewed upon request for access: https://drive.google.com/drive/folders/1ymg1rdiIAARpNazJzKn5KDoCeVx_Skdp Code availability All code used to conduct data preprocessing, simulate data, perform power analyses, and analyze both pilot and final datasets will be made publicly available upon acceptance of the Stage 2 manuscript. The code is currently hosted on a private repository for peer review and will be released under an open-access license following publication. Results A cumulative total of 200 (AD 100 and HC 100) among the Caucasian dataset helped ensure that the study possesses the power to detect significant variations in the prevalence of AD among individuals aged 61 to 90 years. It is essential to guarantee that the study is sufficiently powered to identify significant differences in AD prevalence within the designated age range. Table 1 Age distribution and neuropsychological test scores based on AD and HC group in the Caucasian population. HC AD p-value N (M/F) 45/55 53/47 Age Range (mean (SD)) 61–90 (78.23 (6.093)) 60–90 77.53 (7.317) < 0.001 MMSE 28.52 (1.867) 8.23 (4.390) < 0.001 MMSE = Mini-Mental State Examination; Value = Mean (standard deviation); Significant value p < 0.001 Table 1 presents the age distribution and neuropsychological test scores according to the mean values and standard deviations of the AD and HC groups in the Caucasian population. The mean age for HC and AD groups is 78.23 years and 77.53 years, respectively. The low MMSE scores among the AD patients signify that their cognitive performance is markedly deteriorated in comparison to our HC subjects. Table 2 Age distribution and neuropsychological test scores based on AD and HC group in the Southeast Asian data HC AD p-value N (M/F) 5/10 6/9 AgeRange (mean (SD)) 60–90 (71.2 (7.68)) 60–90 (74.27 (10.55)) < 0.001 MMSE 27.53 (2.61) 14.4 (8.33) < 0.001 Figures 1 and 2 illustrate scatter plot pairs of numerical data depicting the distribution of hippocampus volume according to age for the VBM and HippoDeep-based approaches, contrasting the distributions of the AD and HC groups to elucidate their relationship. The linear graph illustrates the trend of hippocampal volume reduction with advancing age, as this variable is inversely correlated with age. Moreover, the volumes of the right and left hippocampi, assessed using both methodologies, indicated that these values were diminished (demonstrating faster atrophy) in the Alzheimer's disease group relative to the age-matched healthy control group. The ROC curve demonstrates the sensitivity and specificity of each model in differentiating AD from HC, visually representing the trade-off between the true positive rate and false positive rate across various threshold values. Table 3 Comparison of the sensitivity and specificity of HippoDeep versus VBM in determining AD by having a specific cut-off point for hippocampal volume. Parameter Lt HippoDeep Rt HippoDeep Lt VBM Rt VBM Specificity (%) 89 91 94 100 Sensitivity (%) 83 70 48 35 AUC (ROC) 0.918 0.880 0.721 0.687 Associated criterion ≤ 2.31 ≤ 2.26 ≤ 1.73 ≤ 1.48 HippoDeep demonstrated significantly stronger correlations with MMSE scores (left hippocampus: r = 0.65; right: r = 0.59; p < 0.001) compared to VBM (left: r = 0.41; right: r = 0.38; p < 0.05). Mean hippocampal volumes in AD patients were significantly reduced relative to healthy controls across both methods (HippoDeep: p < 0.001; VBM: p < 0.05). AUC: area under the curve Receiver Operating Characteristic (ROC) analysis showed that HippoDeep outperformed VBM in diagnostic accuracy: HippoDeep: AUC = 0.918 (left), 0.882 (right); sensitivity = 83%, specificity = 89% (left). VBM: AUC = 0.788 (left), 0.741 (right); sensitivity = 70%, specificity = 75%. Cut-off thresholds for AD classification were established at ≤ 2.31 cm³ (left) and ≤ 2.26 cm³ (right) using HippoDeep. Cross-cohort comparison revealed consistent trends in hippocampal atrophy across Caucasian and Southeast Asian groups, indicating model generalizability. Discussion The findings of this study underscore the superior performance of the HippoDeep deep learning algorithm over traditional VBM techniques in accurately quantifying hippocampal volumes, which can help to support the diagnosis of AD. Our results demonstrate that hippocampal volumes derived using HippoDeep exhibit a significantly stronger correlation with MMSE scores than those obtained via VBM. This reinforces the clinical relevance of HippoDeep in assessing cognitive decline and tracking disease progression. ROC analysis further supports HippoDeep’s diagnostic utility, revealing higher sensitivity and specificity compared to VBM, particularly in the left hippocampus (sensitivity: 83%, specificity: 89%, AUC: 0.918). The establishment of population-specific volumetric thresholds (≤ 2.31 cm³ for the left and ≤ 2.26 cm³ for the right hippocampus) using the HippoDeep method further enhances the algorithm’s discriminative power, offering an objective and quantitative biomarker for early-stage AD detection. These results position HippoDeep as a valuable tool for clinical decision-making, especially when early intervention is critical. Importantly, our cross-cohort analysis revealed consistent trends in hippocampal atrophy between the Southeast Asian and Caucasian datasets, demonstrating the algorithm’s robustness across diverse ethnic populations. While previous studies have expressed concerns about the potential overfitting of deep learning models in small, demographically constrained datasets—particularly among Southeast Asian populations—our results show comparable performance across both cohorts. This suggests that HippoDeep possesses strong generalizability, although further validation in larger, more heterogeneous Asian populations is warranted to optimize ethnic- and region-specific cut-off thresholds. In comparison to other convolutional neural network (CNN)-based segmentation methods, HippoDeep demonstrates several advantages, including a relatively simplified pipeline, rapid processing times, and reduced need for extensive computational resources. Other advanced CNN architectures such as U-Net variants and 3D convolutional models may yield higher segmentation precision, especially in high-resolution volumetric MRI data; however, they often require significant GPU capacity, longer training times, and more complex preprocessing protocols(Nobakht et al., 2021 ). While these advanced models may achieve higher Dice similarity coefficients—sometimes exceeding 0.90—HippoDeep maintains competitive performance with Dice indices typically ranging between 0.85 and 0.89, offering a strong balance between accuracy and accessibility. Nonetheless, the continued development of lightweight, high-accuracy CNN models remains essential for improving segmentation fidelity in clinical settings with limited computational infrastructure. The observed correlation between hippocampal volume and MMSE scores emphasizes the value of integrating automated neuroimaging tools with cognitive assessments. Such integration can enhance diagnostic accuracy and allow for more personalized monitoring of neurodegenerative progression (Song et al., 2025 ). As deep learning algorithms continue to mature, tools like HippoDeep could significantly streamline clinical workflows, offering rapid and reliable assessments with minimal human input. The relationship between hippocampal volume and MMSE scores underscores the value of integrating automated imaging tools with neuropsychological assessments. As deep learning-based segmentation methods continue to evolve, tools like HippoDeep could significantly streamline MRI interpretation in clinical workflows, particularly in resource-limited settings where radiological expertise may be scarce. Importantly, this study also addresses the practical challenges in widespread clinical implementation of MRI-based assessments in under-resourced regions, where limited access to imaging infrastructure and trained personnel has hindered early diagnosis efforts. HippoDeep’s minimal manual input requirement, combined with high reproducibility and diagnostic accuracy, makes it a promising tool for scalable and standardized hippocampal assessment (Chau et al., 2025 ). We also addressed the key practical challenges for implementing MRI-based diagnostics in developing countries in Southeast Asia, which include scares expertise in AI-based softwares. This often hinders the effort to make an early diagnosis of AD in these regions. HippoDeep’s automated pipeline, which is an open source software, requires minimal manual intervention and provides high reproducibility. This makes it a promising solution for scalable, standardized hippocampal assessment in routine clinical use—even in relatively low-income regions. Furthermore, local researchers in the Asian continent can learn and improvise by studying other work in progress to further improve CNN-based methods for automated measurement of hippocampal method such as the CNN-based multi-model methods for jointly learning hippocampal segmentation and disease classification developed by Liu et al. (Liu et al., 2020). In conclusion, HippoDeep markedly outperforms conventional VBM methods in both segmentation accuracy and diagnostic performance. Its high sensitivity, specificity, and generalizability suggest that it may serve as a reliable and scalable tool for early AD detection across global populations. In comparing datasets across demographics, our analysis revealed comparable trends in hippocampal atrophy between the Southeast Asian and Caucasian datasets. This supports HippoDeep’s generalizability across diverse ethnic backgrounds. We recommend that future studies need to validate the clinical deployment of utilizing HippoDeep in larger, multi-ethnic samples, particularly in Asia, to optimize demographic-specific thresholds and support widespread adoption and equitable care. Prospective trials should be performed to recruit more homogeneous subjects and perform longitudinal follow-up to fully utilize CNN-based algorithms. Declarations Competing interests The authors declare no competing interests. Funding This research was funded by the Ministry of Higher Education, Malaysia under the Fundamental Research Grant Scheme (FRGS/1/2019/SKK03/UPM/02/4) and project code: 04-01-19-2119FR and project number 5540244 that was awarded to Professor Dr. Subapriya Suppiah. The funders have/had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Author Contribution 1.Conceptualization - SS2.Data Curation – NSN, NN3.Formal Analysis - NSN4.Funding Acquisition - SS5.Investigation – NSN, HMA6.Methodology – NSN, HMA, BI, VPS7.Project Administration – NSN, SS8.Resources – NSN, SS, RM, NHH9.Software - NSN10.Supervision – SS, MM, MH, HMS11.Validation - SS, MM12.Visualization - NSN, NN13.Writing – Original Draft - NSN14.Writing – Review & Editing – all authors Acknowledgements We acknowledge the support from the Ministry of Higher Education, Malaysia and the support from the Director General, Ministry of Health Malaysia for the completion of this project. Data Availability We are committed to share our raw data and materials on acceptance of our Stage 2 manuscript. Our data is deposited in a self-repository (pending complete institutional repository upload) and is available to be viewed upon request for access: https://drive.google.com/drive/folders/1ymg1rdiIAARpNazJzKn5KDoCeVx_Skdp References Aboutorab, H., Hussain, O. K., Saberi, M., Hussain, F. K. & Chang, E. A survey on the suitability of risk identification techniques in the current networked environment. J. Netw. Comput. Appl. 178 , 102984. https://doi.org/https://doi.org/10.1016/j.jnca.2021.102984 (2021). Alzheimer's Disease International. Numbers of people with dementia worldwide. (2020)., November 30 https://www.alzint.org/resource/numbers-of-people-with-dementia-worldwide/ Bolchini, C., Cassano, L. & Miele, A. Resilience of Deep Learning applications: a systematic literature review of analysis and hardening techniques . (2023). http://arxiv.org/abs/2309.16733 Chau, M., Vu, H., Debnath, T. & Rahman, M. G. A scoping review of automatic and semi-automatic MRI segmentation in human brain imaging. In Radiography (Vol. 31, Issue 2). W.B. Saunders Ltd. https://doi.org/10.1016/j.radi.2025.01.013 (2025). Di, X. & Biswal, B. B. A functional MRI pre-processing and quality control protocol based on statistical parametric mapping (SPM) and MATLAB. Frontiers in Neuroimaging , 1 . https://www.frontiersin.org/journals/neuroimaging/articles/ (2023). 10.3389/fnimg.2022.1070151 Gaser, C. et al. CAT: a computational anatomy toolbox for the analysis of structural MRI data. GigaScience 13 , giae049. https://doi.org/10.1093/gigascience/giae049 (2024). Gemmell, E. et al. Hippocampal Neuronal Atrophy and Cognitive Function in Delayed Poststroke and Aging-Related Dementias . (2011). https://doi.org/10.1161/STROKEAHA.111.636498/-/DC1 Huang, H. et al. Voxel-based morphometry and a deep learning model for the diagnosis of early Alzheimer’s disease based on cerebral gray matter changes. Cereb. Cortex . 33 (3), 754–763. https://doi.org/10.1093/cercor/bhac099 (2023). Ibrahim, N. S. N. et al. Comparison of deep learning convolutional neural networks method with conventional volume-based morphometry measurement of hippocampal volume in Alzheimer’s disease. Neurosci. Res. Notes . 6 (4). https://doi.org/10.31117/neuroscirn.v6i4.248 (2023). Khagi, B. et al. Vbm-based alzheimer’s disease detection from the region of interest of t1 mri with supportive gaussian smoothing and a bayesian regularized neural network. Appl. Sci. (Switzerland) . 11 (13). https://doi.org/10.3390/app11136175 (2021). Korbmacher, M., Westlye, L. T. & Maximov, I. I. FreeSurfer version-shuffling can enhance brain age predictions. NeuroImage: Rep. 4 (3), 100214. https://doi.org/https://doi.org/10.1016/j.ynirp.2024.100214 (2024). Lavielle, A. et al. T1 Mapping From MPRAGE Acquisitions: Application to the Measurement of the Concentration of Nanoparticles in Tumors for Theranostic Use T1 Mapping From MPRAGE Acquisitions: Application to the Measurement of the Concentration of Nanoparticles in Tumors for T 1 Mapping From MPRAGE Acquisitions: Application to the Measurement of the Concentration of Nanoparticles in Tumors for Theranostic Use. J. Magn. Reson. Imaging . 2023 (1), 313–323. https://doi.org/10.1002/jmri.28509ï (2022). Leow, Y. J. et al. Biomarkers and Cognition Study, Singapore (BIOCIS): Protocol, Study Design, and Preliminary Findings. J. Prev. Alzheimer’s Disease . 11 (4), 1093–1105. https://doi.org/10.14283/jpad.2024.89 (2024). Liu, Liu, M. et al. A multi-model deep convolutional neural network for automatic hippocampus segmentation and classification in Alzheimer’s disease. Neuroimage 208 , 116459. https://doi.org/10.1016/j.neuroimage.2019.116459 (2020). Loftus, J. R., Puri, S. & Meyers, S. P. Multimodality imaging of neurodegenerative disorders with a focus on multiparametric magnetic resonance and molecular imaging. Insights into Imaging . 14 (1). https://doi.org/10.1186/s13244-022-01358-6 (2023). Mostafa, A. M., Zakariah, M. & Aldakheel, E. A. Brain Tumor Segmentation Using Deep Learning on MRI Images. Diagnostics 13 (9). https://doi.org/10.3390/diagnostics13091562 (2023). Nobakht, S. et al. Combined atlas and convolutional neural network-based segmentation of the hippocampus from mri according to the adni harmonized protocol. Sensors 21 (7). https://doi.org/10.3390/s21072427 (2021). Raji, C. A., Meysami, S., Porter, V. R., Merrill, D. A. & Mendez, M. F. Diagnostic utility of brain MRI volumetry in comparing traumatic brain injury, Alzheimer disease and behavioral variant frontotemporal dementia. BMC Neurol. 24 (1). https://doi.org/10.1186/s12883-024-03844-4 (2024). Rao, R., Paramasivam, G., Ramachandra Rao, I. & Prabhu, M. A. Normality Testing in Statistics: What Clinician-Researchers Should Know . (2024). https://doi.org/10.4103/HFJI.HFJI_7_24 Schöll, M. et al. Challenges in the practical implementation of blood biomarkers for Alzheimer’s disease. Lancet Healthy Longev. https://doi.org/10.1016/j.lanhl.2024.07.013 (2024). Sghirripa, S. et al. Evaluating Hippocampal Segmentation Methods Evaluating Traditional, Deep Learning, and Subfield Methods for Automatically Segmenting the Hippocampus from MRI. Hum Brain Mapp. 1;46(5): e70200. (2024). https://doi.org/10.1101/2024.08.06.24311530 Song, J. et al. Development of Neurodegenerative Disease Diagnosis and Monitoring from Traditional to Digital Biomarkers. In Biosensors (Vol. 15, Issue 2). Multidisciplinary Digital Publishing Institute (MDPI). (2025). https://doi.org/10.3390/bios15020102 Subramaniam, M. et al. Prevalence of dementia in Singapore: Changes across a decade. Alzheimer’s Dement. https://doi.org/10.1002/alz.14485 (2025). Sui, C. et al. Decreased gray matter volume in the right middle temporal gyrus associated with cognitive dysfunction in preeclampsia superimposed on chronic hypertension. Front. NeuroSci. 17 , 1138952. https://doi.org/10.3389/fnins.2023.1138952 (2023). Tavares, V., Prata, D. & Ferreira, H. A. Comparing SPM12 and CAT12 segmentation pipelines: a brain tissue volume-based age and Alzheimer’s disease study. J. Neurosci. Methods . 334 , 108565. https://doi.org/https://doi.org/10.1016/j.jneumeth.2019.108565 (2020). Thyreau, B., Sato, K., Fukuda, H. & Taki, Y. Segmentation of the hippocampus by transferring algorithmic knowledge for large cohort processing. Med. Image. Anal. 43 , 214–228. https://doi.org/https://doi.org/10.1016/j.media.2017.11.004 (2018). Vejandla, B., Savani, S., Appalaneni, R., Veeravalli, R. S. & Gude, S. S. Alzheimer’s Disease: The Past, Present, and Future of a Globally Progressive Disease. Cureus . (2024). https://doi.org/10.7759/cureus.51705 Xing, K., Ku, J. & Zhao, J. A Novel Approach to Optimizing Convolutional Neural Networks for Improved Digital Image Segmentation. Int. J. Intell. Syst. 2024 (1), 4337255. https://doi.org/https://doi.org/10.1155/2024/4337255 (2024). Zhao, Z., Alzubaidi, L., Zhang, J., Duan, Y. & Gu, Y. A comparison review of transfer learning and self-supervised learning: Definitions, applications, advantages and limitations. Expert Syst. Appl. 242 , 122807. https://doi.org/https://doi.org/10.1016/j.eswa.2023.122807 (2024). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 02 Dec, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 25 Aug, 2025 Reviews received at journal 23 Aug, 2025 Reviews received at journal 12 Aug, 2025 Reviewers agreed at journal 12 Aug, 2025 Reviewers agreed at journal 07 Aug, 2025 Reviewers agreed at journal 04 Aug, 2025 Reviewers agreed at journal 04 Aug, 2025 Reviewers invited by journal 25 Jul, 2025 Editor assigned by journal 17 Jun, 2025 Submission checks completed at journal 09 Jun, 2025 First submitted to journal 09 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6790322","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":491767561,"identity":"78ad1bad-0ee0-470a-8924-ec235f6e87e2","order_by":0,"name":"Nur Shahidatul Nabila Ibrahim","email":"","orcid":"","institution":"Universiti Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Nur","middleName":"Shahidatul Nabila","lastName":"Ibrahim","suffix":""},{"id":491767562,"identity":"4d9400b9-58f7-47df-8baf-e65e0dba8688","order_by":1,"name":"Subapriya Suppiah","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5klEQVRIiWNgGAWjYFACHhAhwcDA3gDmMjYQr4XnAGlaQLoSiNQiPyP34OfCPRaJ/TPfmD26wWAju+EAd5oEPi0GN/KSpWc8k0iccTvH3DiHIc14wwHebfi1SOQYSPMckEhsuJ1jJp3DcDiRoBb5GTnGv0Fa5t88A9Lyn7AWhhtAw0FaNtzgAWk5QFiLwZl3adYzDkgYbzyTVm6cY5BsPPMw72YLvA5rzz18u+BAney844e3Pc6psJPtO9678QZehwEBMxA7NjAwsAEtBXNZ8PsFqsWeAawFKvCBkJZRMApGwSgYUQAAkahMp/0dsswAAAAASUVORK5CYII=","orcid":"","institution":"Universiti Putra Malaysia","correspondingAuthor":true,"prefix":"","firstName":"Subapriya","middleName":"","lastName":"Suppiah","suffix":""},{"id":491767563,"identity":"c048f6d2-98a7-4e62-ae60-eb7a414cddb5","order_by":2,"name":"Buhari Ibrahim","email":"","orcid":"","institution":"Universiti Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Buhari","middleName":"","lastName":"Ibrahim","suffix":""},{"id":491767564,"identity":"0e4230fb-23b4-473d-81e9-7cbc9d3b362c","order_by":3,"name":"Nur Hafizah Mohad Azmi","email":"","orcid":"","institution":"Universiti Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Nur","middleName":"Hafizah Mohad","lastName":"Azmi","suffix":""},{"id":491767567,"identity":"8ba415c7-5ec5-4a63-b137-c1eb279f1ed5","order_by":4,"name":"Vengkatha Priya Seriramulu","email":"","orcid":"","institution":"Universiti Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Vengkatha","middleName":"Priya","lastName":"Seriramulu","suffix":""},{"id":491767569,"identity":"01d4df29-3f2f-41de-bce7-3726d3863848","order_by":5,"name":"Mazlyfarina Mohamad","email":"","orcid":"","institution":"Universiti Kebangsaan Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Mazlyfarina","middleName":"","lastName":"Mohamad","suffix":""},{"id":491767570,"identity":"a3ff7d41-c8ea-499e-8df9-95ad244e471f","order_by":6,"name":"Marsyita Hanafi","email":"","orcid":"","institution":"Universiti Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Marsyita","middleName":"","lastName":"Hanafi","suffix":""},{"id":491767571,"identity":"12f96586-e409-4862-8ee6-0d28a46024c2","order_by":7,"name":"Hakimah Mohammad Sallehuddin","email":"","orcid":"","institution":"Hospital Sultan Abdul Aziz Shah, Universiti Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Hakimah","middleName":"Mohammad","lastName":"Sallehuddin","suffix":""},{"id":491767572,"identity":"74dfd502-8595-4fba-a1cb-c1be25a40fc5","order_by":8,"name":"Nurallysha Najwa","email":"","orcid":"","institution":"KPJ Healthcare University","correspondingAuthor":false,"prefix":"","firstName":"Nurallysha","middleName":"","lastName":"Najwa","suffix":""},{"id":491767575,"identity":"c610ac4e-3c57-4da6-a897-98e8bb3df7e6","order_by":9,"name":"Rizah Mazzuin Razali","email":"","orcid":"","institution":"Jabatan Perubatan, Hospital Kuala Lumpur","correspondingAuthor":false,"prefix":"","firstName":"Rizah","middleName":"Mazzuin","lastName":"Razali","suffix":""},{"id":491767576,"identity":"7a03a8e6-c9e0-46bf-8c51-9d771e75ba34","order_by":10,"name":"Noor Harzana Harrun","email":"","orcid":"","institution":"Klinik Kesihatan Pandamaran","correspondingAuthor":false,"prefix":"","firstName":"Noor","middleName":"Harzana","lastName":"Harrun","suffix":""}],"badges":[],"createdAt":"2025-05-31 10:23:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6790322/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6790322/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-29366-8","type":"published","date":"2025-12-02T15:57:25+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":87882230,"identity":"db1fa413-ba9c-498b-a93b-bb3b0c5e20de","added_by":"auto","created_at":"2025-07-30 04:30:38","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":353943,"visible":true,"origin":"","legend":"\u003cp\u003eRight hippocampal volume distribution according to age different between Caucasian and Southeast Asian dataset analysis for AD compared to HC groups.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6790322/v1/b093841ad3a2b21fee1143b7.jpeg"},{"id":87882853,"identity":"dc8814f4-77d6-44af-acf7-3d65a67e38c8","added_by":"auto","created_at":"2025-07-30 04:46:38","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":412750,"visible":true,"origin":"","legend":"\u003cp\u003eLeft hippocampal volume distribution according to age different between Caucasian and Southeast Asian dataset analysis for AD compared to HC groups.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6790322/v1/fc9f92095d9120453c51b7e6.jpeg"},{"id":87882229,"identity":"b3938d69-8f00-4a70-90a0-88586d1607b8","added_by":"auto","created_at":"2025-07-30 04:30:38","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":258417,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve showing the diagnostic accuracy of the hippocampal volume assessment methods. (HippoDeep: Right hippocampus; HippoDeep: Left hippocampus; VBM: Left hippocampus, and VBM: Reft hippocampus)\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6790322/v1/5d714b0b7ec0a05478bc3295.png"},{"id":97723815,"identity":"41407360-7417-45d6-bc98-2c92d298e6d2","added_by":"auto","created_at":"2025-12-08 16:07:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1702981,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6790322/v1/a759a820-ed1b-4457-b833-a97357591b3f.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comparing Deep Learning CNN method with Traditional MRI-based Hippocampal segmentation and volumetry for Early Alzheimer’s Disease Diagnosis Across Diverse Populations","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAlzheimer\u0026rsquo;s disease (AD) and other dementias have progressively become more prevalent in Southern Asia due to the growing elderly population, which has placed a greater burden on the healthcare systems (Zhu et al., 2023). Research reveals that over 3.55\u0026nbsp;million individuals in Southeast Asia are afflicted with dementia, accounting for 8.5% of older adults in Malaysia living with AD, i.e. approximately 260,000 people and projected to rise to 590,000 by 2050 (Alzheimer\u0026rsquo;s Disease International, 2020). According to Alzheimer's Disease International (ADI), the number of individuals with dementia in Indonesia is expected to reach nearly 4\u0026nbsp;million by 2050 (Leow et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The need to monitor the trends in the prevalence of dementia as populations age with increasing life expectancy is paramount for health care planning and resource allocation in Singapore (Subramaniam et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, there has been no standard monitoring of this population. Moreover, insufficient awareness of AD has caused it to become the subject of stigma and misunderstanding among the general public, which may impede early detection and therapy.\u003c/p\u003e\u003cp\u003eAD is a neurodegenerative condition that is characterised by progressive cognitive decline and distinctive symptoms like brain shrinkage and beta-amyloid accumulation (Raji et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Hippocampal atrophy is a hallmark biomarker of AD and other forms of dementia, as well as cerebrovascular accidents (CVA), underscoring its clinical relevance in neurodegenerative and vascular pathologies (Gemmell et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Vejandla et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Conversely, a correlation was discovered between PET-Aβ and overall hippocampal atrophy in cognitively intact elderly persons (Xia et al., 2024). Aβ has a more pronounced impact on hippocampus degeneration in the initial phases of the disease (Rao et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), whereas other conditions contribute more significantly at the later stages of dementia. Particularly in normal aging, the hippocampal volume decreases at a gradual rate, approximately 1\u0026ndash;2% per year after the age of 60. This mild shrinkage is typically associated with age-related cognitive decline but does not significantly impair daily functioning. In contrast, AD is characterized by accelerated hippocampal atrophy, with rates ranging from 3\u0026ndash;5% per year in the early stages and potentially higher as the disease progresses. This rapid loss is strongly correlated with declining episodic memory and cognitive performance, distinguishing AD from normal aging (Sghirripa et al., 2025).\u003c/p\u003e\u003cp\u003eTraditionally, structural magnetic resonance imaging (sMRI) is the preferred non-invasive detection tool to identify the features of AD during the the later stages, however, this can be challenging in the early phase, despite \u003cem\u003ea priori\u003c/em\u003e knowledge of hippocampal atrophy and cortical thinning being associated with cognitive decline in AD (Mostafa et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). T1 MPRAGE MRI scans, in the coronal orthogonal plane, provide comprehensive visualisation of structural alterations in the brain, enabling precise evaluation and monitoring of disease progression (Lavielle et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This imaging technique is crucial for the early detection and observation of changes in the brain related to AD. Quantitative sMRI can forecast the progression to dementia in individuals with moderate cognitive impairment. Moreover, quantitative sMRI can yield critical insights for prediction of disease and also help dictate strategies for early interventions. Integrating this imaging technology into clinical practice can enhance diagnostic precision and patient outcomes for those who are predisposed to dementia (Loftus et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Given the strong association between hippocampal volume reduction and cognitive impairment, accurate segmentation of the hippocampus using sMRI is crucial for early diagnosis, monitoring, and research applications. While manual segmentation remains the gold standard due to its anatomical precision, it is time-consuming, operator-dependent, and often lacks reproducibility across raters\u003c/p\u003e\u003cp\u003eVoxel-based morphometry (VBM) is a semi-automated neuroimaging methodology employed to evaluate structural variations in the brain, with has encompassed alterations in the hippocampus volume. Although VBM is proficient for group-level analyses, it is less precise at the individual level in comparison to automated segmentation techniques. Fully automated segmentation methods using deep learning (DL) algorithms such as convolutional neural networks (CNN), have led to the development of various softwares. Multiple automated hippocampal segmentation methods have emerged over the past two decades. These include atlas-based techniques such as FreeSurfer, which have been extensively validated but show limitations in populations with brain atrophy or lesions, such as those following CVA. A significant drawback of many of these algorithms is their reliance on young, healthy brain templates, lack of generalizability to aging and diseased brains, and operational demands in terms of computational resources and user expertise. Initial studies(Thyreau et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) have shown that HippoDeep, which was developed using FreeSurfer, has been used successfully in non-demented (non-AD) subjects, providing reliable hippocampal volume assessments. The efficacy in bigger AD populations has yet to be comprehensively evaluated. Nevertheless, HippoDeep's capacity to efficiently process extensive datasets renders it advantageous for wider applicability in AD research. Additional validation in Alzheimer's disease-specific cohorts is essential to confirm its diagnostic utility.\u003c/p\u003e\u003cp\u003eWhile HippoDeep has been successfully employed for automated hippocampal segmentation and measurement using normative dataset of stroke and AD patients from the Western populations (Sghirripa et al., 2025), its performance among AD subjects in an Asian population dataset has not been comprehensively assessed. This may be due to a lack of information regarding the availability of imaging technologies, an inadequacy of training skills for interpretation, and the requirement of high computational resources for the evaluations of neurodegenerative disorders (Sch\u0026ouml;ll et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Hence, HippoDeep is recommended as the preferred CNN-based automated hippocampal segmentation and volumetry method that\u0026rsquo;s helps to reduce the interpretation time and enhance the standardization of brain measurements for better longitudinal comparison.\u003c/p\u003e\u003cp\u003eDespite their theoretical advantages, CNN-based segmentation methods such as HippoDeep have yet to be widely validated in non-Western populations and under clinical conditions common in resource-constrained settings. We aimed to determine the difference in the range of hippocampal volume in a Southeast Asian AD population dataset, comprising of people from an Austronesian group indigenous to the Malay Peninsula, compared with dataset from a Caucasian population from the ADNI database. We also evaluated the diagnostic accuracy of the CNN-based HippoDeep segmentation model with traditional VBM method in assessing hippocampal volume among AD patients and healthy controls (HC) in a Caucasian compared to a Southeast Asian population, by establishing hippocampal volume cut-off thresholds, using ROC curve analysis. Subsequently, we assessed the generalizability and diagnostic performance of HippoDeep in a diverse population. Finally, we explored the relationship between hippocampal volume loss and cognitive impairment, as measured by Mini-Mental State Examination (MMSE) scores, to validate the clinical relevance of this CNN-based automated segmentation tool.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eEthics information\u003c/h2\u003e\u003cp\u003e This study was approved by the Malaysian national research ethics committee, which is the Medical Research Ethics Committee (MREC) of National Medical Registration Registry (NMRR) Malaysia (NMRR-19-2719-49105) and the Ethics Committee for Research Involving Human Subjects of Universiti Putra Malaysia (JKEUPM-2019-328). All the subjects from this nationally conducted research gave their informed consent, which was obtained from the participants and/or their legal guardians. Furthermore, the research was performed in accordance with the Declaration of Helsinki.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eDesign\u003c/h3\u003e\n\u003cp\u003eThis study utilized a dataset comprising of Caucasian subjects from the Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu) and also data of Southeast Asian subjects from a national Fundamental Research Grant Scheme (FRGS) project in Malaysia. Participants diagnosed with AD were identified based on clinical criteria and underwent multiple neuroimaging evaluations, allowing for a comprehensive analysis of disease progression and biomarker changes over time. Neuropsychological assessment included a range of cognitive evaluation tools, such as the Mini-Mental State Examination (MMSE). Cognitively healthy control (HC) subjects in the Southeast Asian study were recruited through advertisements on community bulletin boards and distributed fliers. The dataset for Southeast Asian subjects is in the process of being deposited in our academic institutional repository.\u003c/p\u003e\u003cp\u003eWe evaluated the performance of hippocampal segmentation using both deep learning (DL)-based and traditional atlas-based methods on sMRI scans from two demographically distinct datasets: the Alzheimer's Disease Neuroimaging Initiative (ADNI) and a Southeast Asian FRGS study memory clinic cohort. CNN-based segmentation was conducted using HippoDeep, while VBM method using SPM Mathlab represented the traditional atlas-based approach.\u003c/p\u003e\u003cp\u003eThe ADNI database solely provides gender and age demographic data. Age is continuous, while gender is female or male. The findings may not apply to larger populations due to this constraint. For analysis, data pre-processing standardised datasets from multiple sources and imaging data. Pre-processing neuroimaging data corrected head motion and standardised picture intensity across sessions. The Dice Similarity Coefficient (DSC) was used to compare automated and manual segmentations on 30 representative images from each cohort to assess segmentation quality. Statistical analysis includes ANOVA for group comparisons and Pearson correlation for hippocampus size and MMSE scores.\u003c/p\u003e\u003cp\u003eThis study compared bilateral hippocampus volume measurements from VBM and HippoDeep. Additionally, we examined the link between cognitive performance evaluations and hippocampal volume. HippoDeep and VBM were evaluated for AD diagnosis accuracy using ROC curve with the MMSE scores combined with clinical assessment criteria as a reference standard.\u003c/p\u003e\n\u003ch3\u003eSampling plan\u003c/h3\u003e\n\u003cp\u003eThe methodology employed for hippocampal volume segmentation in this study was derived from prior research (Thyreau et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The authors relied on the availability of FreeSurfer (RRID:SCR_001847) software and pre-existing FreeSurfer-labeled online datasets as a crucial knowledge store. This facilitates the swift and significant enhancement of training examples (Korbmacher et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Transfer learning is a machine learning approach that involves training a model using the results of a sophisticated algorithm (Aboutorab et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In deep learning, transfer learning typically refers to the fine-tuning of a model that has been pre-trained on a larger, more diverse dataset, which is then utilised in a more specialised area with restricted datasets (Zhao et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The researchers utilised simulated pictures of arterial trees produced by a physiologically accurate model to get a high level of precision and robustness (Bolchini et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Xing et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The simulated photos functioned as the input for the classifier's training and evaluation.\u003c/p\u003e\u003cp\u003eThe installation of HippoDeep Brain hippocampus segmentation was conducted using a repository compatible with Ubuntu/Linux and Mac OS: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/bthyreau/hippodeep_pytorch\u003c/span\u003e\u003cspan address=\"https://github.com/bthyreau/hippodeep_pytorch\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eWe employed the executable Windows pre-compiled application, which is a plug-and-play application. The pre-compiled application designed for the Windows platform can be accessed at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://drive.usercontent.google.com/download?id=1rSp0S7kCR8ksPiFuujbhbDe0L-qHbmdx\u0026amp;export=download\u0026amp;authuser=0\u003c/span\u003e\u003cspan address=\"https://drive.usercontent.google.com/download?id=1rSp0S7kCR8ksPiFuujbhbDe0L-qHbmdx\u0026amp;export=download\u0026amp;authuser=0\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (Ibrahim et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe VBM procedure(Huang et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) initiated with segmentation (spatial pre-processing): employing the native space option, a tissue class image aligned with the original was generated, subsequently followed by the implementation of Dartel(Sui et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) to produce templates from selected images for distortion correction. The next process was normalisation, during which the tissue class image was transformed into a standardised space, followed by smoothing to improve the signal-to-noise ratio. These techniques are crucial for data preparation before subsequent analysis. Imagesgit with similar dimensions, orientation, and voxel size were selected for statistical analysis using a factorial design specification. The value was calculated via a design specification file from the computer's file system. The Automated Anatomical Labelling (AAL) toolkit(Khagi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) was superimposed onto the Montreal Neurological Institute (MNI) template(Gaser et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) to derive the volume at the cluster level.\u003c/p\u003e\n\u003ch3\u003eAnalysis Plan\u003c/h3\u003e\n\u003cp\u003eAll the T1-weighted images were preprocessed using SPM Toolbox (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.fil.ion.ucl.ac.uk/spm/download/restricted/eldorado/spm12.zip\u003c/span\u003e\u003cspan address=\"https://www.fil.ion.ucl.ac.uk/spm/download/restricted/eldorado/spm12.zip\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) implemented in MATLAB (R2021a) (Di \u0026amp; Biswal, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Tavares et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eData availability\u003c/h3\u003e\n\u003cp\u003eWe are committed to share our raw data and materials on acceptance of our Stage 2 manuscript. Our data is deposited in a self-repository (pending complete institutional repository upload) and is available to be viewed upon request for access: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://drive.google.com/drive/folders/1ymg1rdiIAARpNazJzKn5KDoCeVx_Skdp\u003c/span\u003e\u003cspan address=\"https://drive.google.com/drive/folders/1ymg1rdiIAARpNazJzKn5KDoCeVx_Skdp\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eCode availability\u003c/h2\u003e\u003cp\u003eAll code used to conduct data preprocessing, simulate data, perform power analyses, and analyze both pilot and final datasets will be made publicly available upon acceptance of the Stage 2 manuscript. The code is currently hosted on a private repository for peer review and will be released under an open-access license following publication.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eA cumulative total of 200 (AD 100 and HC 100) among the Caucasian dataset helped ensure that the study possesses the power to detect significant variations in the prevalence of AD among individuals aged 61 to 90 years. It is essential to guarantee that the study is sufficiently powered to identify significant differences in AD prevalence within the designated age range.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAge distribution and neuropsychological test scores based on AD and HC group in the Caucasian population.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHC\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAD\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003ep-value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eN (M/F)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e45/55\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e53/47\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge Range\u003c/p\u003e\u003cp\u003e(mean (SD))\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e61\u0026ndash;90\u003c/p\u003e\u003cp\u003e(78.23 (6.093))\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e60\u0026ndash;90\u003c/p\u003e\u003cp\u003e77.53 (7.317)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMMSE\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e28.52 (1.867)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e8.23 (4.390)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eMMSE\u0026thinsp;=\u0026thinsp;Mini-Mental State Examination; Value\u0026thinsp;=\u0026thinsp;Mean (standard deviation); Significant value p\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the age distribution and neuropsychological test scores according to the mean values and standard deviations of the AD and HC groups in the Caucasian population. The mean age for HC and AD groups is 78.23 years and 77.53 years, respectively. The low MMSE scores among the AD patients signify that their cognitive performance is markedly deteriorated in comparison to our HC subjects.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAge distribution and neuropsychological test scores based on AD and HC group in the Southeast Asian data\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHC\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAD\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003ep-value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eN (M/F)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5/10\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6/9\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAgeRange\u003c/p\u003e\u003cp\u003e(mean (SD))\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e60\u0026ndash;90\u003c/p\u003e\u003cp\u003e(71.2 (7.68))\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e60\u0026ndash;90\u003c/p\u003e\u003cp\u003e(74.27 (10.55))\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMMSE\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e27.53 (2.61)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e14.4 (8.33)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eFigures \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrate scatter plot pairs of numerical data depicting the distribution of hippocampus volume according to age for the VBM and HippoDeep-based approaches, contrasting the distributions of the AD and HC groups to elucidate their relationship. The linear graph illustrates the trend of hippocampal volume reduction with advancing age, as this variable is inversely correlated with age. Moreover, the volumes of the right and left hippocampi, assessed using both methodologies, indicated that these values were diminished (demonstrating faster atrophy) in the Alzheimer's disease group relative to the age-matched healthy control group.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe ROC curve demonstrates the sensitivity and specificity of each model in differentiating AD from HC, visually representing the trade-off between the true positive rate and false positive rate across various threshold values.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eComparison of the sensitivity and specificity of HippoDeep versus VBM in determining AD by having a specific cut-off point for hippocampal volume.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLt HippoDeep\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRt HippoDeep\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLt VBM\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRt VBM\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpecificity (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e94\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSensitivity (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e83\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e48\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e35\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAUC (ROC)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.918\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.880\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.721\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.687\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAssociated criterion\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u0026le; 2.31\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u0026le; 2.26\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026le; 1.73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u0026le; 1.48\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHippoDeep demonstrated significantly stronger correlations with MMSE scores (left hippocampus: r\u0026thinsp;=\u0026thinsp;0.65; right: r\u0026thinsp;=\u0026thinsp;0.59; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) compared to VBM (left: r\u0026thinsp;=\u0026thinsp;0.41; right: r\u0026thinsp;=\u0026thinsp;0.38; p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Mean hippocampal volumes in AD patients were significantly reduced relative to healthy controls across both methods (HippoDeep: p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; VBM: p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). AUC: area under the curve\u003c/p\u003e\u003cp\u003eReceiver Operating Characteristic (ROC) analysis showed that HippoDeep outperformed VBM in diagnostic accuracy: HippoDeep: AUC\u0026thinsp;=\u0026thinsp;0.918 (left), 0.882 (right); sensitivity\u0026thinsp;=\u0026thinsp;83%, specificity\u0026thinsp;=\u0026thinsp;89% (left). VBM: AUC\u0026thinsp;=\u0026thinsp;0.788 (left), 0.741 (right); sensitivity\u0026thinsp;=\u0026thinsp;70%, specificity\u0026thinsp;=\u0026thinsp;75%.\u003c/p\u003e\u003cp\u003eCut-off thresholds for AD classification were established at \u0026le;\u0026thinsp;2.31 cm\u0026sup3; (left) and \u0026le;\u0026thinsp;2.26 cm\u0026sup3; (right) using HippoDeep. Cross-cohort comparison revealed consistent trends in hippocampal atrophy across Caucasian and Southeast Asian groups, indicating model generalizability.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe findings of this study underscore the superior performance of the HippoDeep deep learning algorithm over traditional VBM techniques in accurately quantifying hippocampal volumes, which can help to support the diagnosis of AD. Our results demonstrate that hippocampal volumes derived using HippoDeep exhibit a significantly stronger correlation with MMSE scores than those obtained via VBM. This reinforces the clinical relevance of HippoDeep in assessing cognitive decline and tracking disease progression.\u003c/p\u003e\u003cp\u003eROC analysis further supports HippoDeep\u0026rsquo;s diagnostic utility, revealing higher sensitivity and specificity compared to VBM, particularly in the left hippocampus (sensitivity: 83%, specificity: 89%, AUC: 0.918). The establishment of population-specific volumetric thresholds (\u0026le;\u0026thinsp;2.31 cm\u0026sup3; for the left and \u0026le;\u0026thinsp;2.26 cm\u0026sup3; for the right hippocampus) using the HippoDeep method further enhances the algorithm\u0026rsquo;s discriminative power, offering an objective and quantitative biomarker for early-stage AD detection. These results position HippoDeep as a valuable tool for clinical decision-making, especially when early intervention is critical.\u003c/p\u003e\u003cp\u003eImportantly, our cross-cohort analysis revealed consistent trends in hippocampal atrophy between the Southeast Asian and Caucasian datasets, demonstrating the algorithm\u0026rsquo;s robustness across diverse ethnic populations. While previous studies have expressed concerns about the potential overfitting of deep learning models in small, demographically constrained datasets\u0026mdash;particularly among Southeast Asian populations\u0026mdash;our results show comparable performance across both cohorts. This suggests that HippoDeep possesses strong generalizability, although further validation in larger, more heterogeneous Asian populations is warranted to optimize ethnic- and region-specific cut-off thresholds.\u003c/p\u003e\u003cp\u003eIn comparison to other convolutional neural network (CNN)-based segmentation methods, HippoDeep demonstrates several advantages, including a relatively simplified pipeline, rapid processing times, and reduced need for extensive computational resources. Other advanced CNN architectures such as U-Net variants and 3D convolutional models may yield higher segmentation precision, especially in high-resolution volumetric MRI data; however, they often require significant GPU capacity, longer training times, and more complex preprocessing protocols(Nobakht et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). While these advanced models may achieve higher Dice similarity coefficients\u0026mdash;sometimes exceeding 0.90\u0026mdash;HippoDeep maintains competitive performance with Dice indices typically ranging between 0.85 and 0.89, offering a strong balance between accuracy and accessibility. Nonetheless, the continued development of lightweight, high-accuracy CNN models remains essential for improving segmentation fidelity in clinical settings with limited computational infrastructure.\u003c/p\u003e\u003cp\u003eThe observed correlation between hippocampal volume and MMSE scores emphasizes the value of integrating automated neuroimaging tools with cognitive assessments. Such integration can enhance diagnostic accuracy and allow for more personalized monitoring of neurodegenerative progression (Song et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). As deep learning algorithms continue to mature, tools like HippoDeep could significantly streamline clinical workflows, offering rapid and reliable assessments with minimal human input. The relationship between hippocampal volume and MMSE scores underscores the value of integrating automated imaging tools with neuropsychological assessments. As deep learning-based segmentation methods continue to evolve, tools like HippoDeep could significantly streamline MRI interpretation in clinical workflows, particularly in resource-limited settings where radiological expertise may be scarce.\u003c/p\u003e\u003cp\u003eImportantly, this study also addresses the practical challenges in widespread clinical implementation of MRI-based assessments in under-resourced regions, where limited access to imaging infrastructure and trained personnel has hindered early diagnosis efforts. HippoDeep\u0026rsquo;s minimal manual input requirement, combined with high reproducibility and diagnostic accuracy, makes it a promising tool for scalable and standardized hippocampal assessment (Chau et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWe also addressed the key practical challenges for implementing MRI-based diagnostics in developing countries in Southeast Asia, which include scares expertise in AI-based softwares. This often hinders the effort to make an early diagnosis of AD in these regions. HippoDeep\u0026rsquo;s automated pipeline, which is an open source software, requires minimal manual intervention and provides high reproducibility. This makes it a promising solution for scalable, standardized hippocampal assessment in routine clinical use\u0026mdash;even in relatively low-income regions. Furthermore, local researchers in the Asian continent can learn and improvise by studying other work in progress to further improve CNN-based methods for automated measurement of hippocampal method such as the CNN-based multi-model methods for jointly learning hippocampal segmentation and disease classification developed by Liu et al. (Liu et al., 2020).\u003c/p\u003e\u003cp\u003eIn conclusion, HippoDeep markedly outperforms conventional VBM methods in both segmentation accuracy and diagnostic performance. Its high sensitivity, specificity, and generalizability suggest that it may serve as a reliable and scalable tool for early AD detection across global populations. In comparing datasets across demographics, our analysis revealed comparable trends in hippocampal atrophy between the Southeast Asian and Caucasian datasets. This supports HippoDeep\u0026rsquo;s generalizability across diverse ethnic backgrounds.\u003c/p\u003e\u003cp\u003eWe recommend that future studies need to validate the clinical deployment of utilizing HippoDeep in larger, multi-ethnic samples, particularly in Asia, to optimize demographic-specific thresholds and support widespread adoption and equitable care. Prospective trials should be performed to recruit more homogeneous subjects and perform longitudinal follow-up to fully utilize CNN-based algorithms.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eCompeting interests\u003c/h2\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis research was funded by the Ministry of Higher Education, Malaysia under the Fundamental Research Grant Scheme (FRGS/1/2019/SKK03/UPM/02/4) and project code: 04-01-19-2119FR and project number 5540244 that was awarded to Professor Dr. Subapriya Suppiah. The funders have/had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003e1.Conceptualization - SS2.Data Curation \u0026ndash; NSN, NN3.Formal Analysis - NSN4.Funding Acquisition - SS5.Investigation \u0026ndash; NSN, HMA6.Methodology \u0026ndash; NSN, HMA, BI, VPS7.Project Administration \u0026ndash; NSN, SS8.Resources \u0026ndash; NSN, SS, RM, NHH9.Software - NSN10.Supervision \u0026ndash; SS, MM, MH, HMS11.Validation - SS, MM12.Visualization - NSN, NN13.Writing \u0026ndash; Original Draft - NSN14.Writing \u0026ndash; Review \u0026amp; Editing \u0026ndash; all authors\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e\u003cp\u003eWe acknowledge the support from the Ministry of Higher Education, Malaysia and the support from the Director General, Ministry of Health Malaysia for the completion of this project.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eWe are committed to share our raw data and materials on acceptance of our Stage 2 manuscript. Our data is deposited in a self-repository (pending complete institutional repository upload) and is available to be viewed upon request for access: https://drive.google.com/drive/folders/1ymg1rdiIAARpNazJzKn5KDoCeVx_Skdp\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAboutorab, H., Hussain, O. K., Saberi, M., Hussain, F. K. \u0026amp; Chang, E. A survey on the suitability of risk identification techniques in the current networked environment. \u003cem\u003eJ. Netw. Comput. Appl.\u003c/em\u003e \u003cb\u003e178\u003c/b\u003e, 102984. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.jnca.2021.102984\u003c/span\u003e\u003cspan address=\"10.1016/j.jnca.2021.102984\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlzheimer's Disease International. Numbers of people with dementia worldwide. (2020)., November 30 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.alzint.org/resource/numbers-of-people-with-dementia-worldwide/\u003c/span\u003e\u003cspan address=\"https://www.alzint.org/resource/numbers-of-people-with-dementia-worldwide/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBolchini, C., Cassano, L. \u0026amp; Miele, A. \u003cem\u003eResilience of Deep Learning applications: a systematic literature review of analysis and hardening techniques\u003c/em\u003e. (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/2309.16733\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/2309.16733\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChau, M., Vu, H., Debnath, T. \u0026amp; Rahman, M. G. A scoping review of automatic and semi-automatic MRI segmentation in human brain imaging. In Radiography (Vol. 31, Issue 2). W.B. Saunders Ltd. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.radi.2025.01.013\u003c/span\u003e\u003cspan address=\"10.1016/j.radi.2025.01.013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDi, X. \u0026amp; Biswal, B. B. A functional MRI pre-processing and quality control protocol based on statistical parametric mapping (SPM) and MATLAB. \u003cem\u003eFrontiers in Neuroimaging\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.frontiersin.org/journals/neuroimaging/articles/\u003c/span\u003e\u003cspan address=\"https://www.frontiersin.org/journals/neuroimaging/articles/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fnimg.2022.1070151\u003c/span\u003e\u003cspan address=\"10.3389/fnimg.2022.1070151\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGaser, C. et al. CAT: a computational anatomy toolbox for the analysis of structural MRI data. \u003cem\u003eGigaScience\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, giae049. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/gigascience/giae049\u003c/span\u003e\u003cspan address=\"10.1093/gigascience/giae049\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGemmell, E. et al. \u003cem\u003eHippocampal Neuronal Atrophy and Cognitive Function in Delayed Poststroke and Aging-Related Dementias\u003c/em\u003e. (2011). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/STROKEAHA.111.636498/-/DC1\u003c/span\u003e\u003cspan address=\"10.1161/STROKEAHA.111.636498/-/DC1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang, H. et al. Voxel-based morphometry and a deep learning model for the diagnosis of early Alzheimer\u0026rsquo;s disease based on cerebral gray matter changes. \u003cem\u003eCereb. Cortex\u003c/em\u003e. \u003cb\u003e33\u003c/b\u003e (3), 754\u0026ndash;763. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/cercor/bhac099\u003c/span\u003e\u003cspan address=\"10.1093/cercor/bhac099\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIbrahim, N. S. N. et al. Comparison of deep learning convolutional neural networks method with conventional volume-based morphometry measurement of hippocampal volume in Alzheimer\u0026rsquo;s disease. \u003cem\u003eNeurosci. Res. Notes\u003c/em\u003e. \u003cb\u003e6\u003c/b\u003e (4). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.31117/neuroscirn.v6i4.248\u003c/span\u003e\u003cspan address=\"10.31117/neuroscirn.v6i4.248\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhagi, B. et al. Vbm-based alzheimer\u0026rsquo;s disease detection from the region of interest of t1 mri with supportive gaussian smoothing and a bayesian regularized neural network. \u003cem\u003eAppl. Sci. (Switzerland)\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e (13). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/app11136175\u003c/span\u003e\u003cspan address=\"10.3390/app11136175\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKorbmacher, M., Westlye, L. T. \u0026amp; Maximov, I. I. FreeSurfer version-shuffling can enhance brain age predictions. \u003cem\u003eNeuroImage: Rep.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e (3), 100214. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.ynirp.2024.100214\u003c/span\u003e\u003cspan address=\"10.1016/j.ynirp.2024.100214\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLavielle, A. et al. T1 Mapping From MPRAGE Acquisitions: Application to the Measurement of the Concentration of Nanoparticles in Tumors for Theranostic Use T1 Mapping From MPRAGE Acquisitions: Application to the Measurement of the Concentration of Nanoparticles in Tumors for T 1 Mapping From MPRAGE Acquisitions: Application to the Measurement of the Concentration of Nanoparticles in Tumors for Theranostic Use. \u003cem\u003eJ. Magn. Reson. Imaging\u003c/em\u003e. \u003cb\u003e2023\u003c/b\u003e (1), 313\u0026ndash;323. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/jmri.28509\u0026iuml;\u003c/span\u003e\u003cspan address=\"10.1002/jmri.28509\u0026iuml;\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLeow, Y. J. et al. Biomarkers and Cognition Study, Singapore (BIOCIS): Protocol, Study Design, and Preliminary Findings. \u003cem\u003eJ. Prev. Alzheimer\u0026rsquo;s Disease\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e (4), 1093\u0026ndash;1105. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.14283/jpad.2024.89\u003c/span\u003e\u003cspan address=\"10.14283/jpad.2024.89\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu, Liu, M. et al. A multi-model deep convolutional neural network for automatic hippocampus segmentation and classification in Alzheimer\u0026rsquo;s disease. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cb\u003e208\u003c/b\u003e, 116459. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neuroimage.2019.116459\u003c/span\u003e\u003cspan address=\"10.1016/j.neuroimage.2019.116459\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLoftus, J. R., Puri, S. \u0026amp; Meyers, S. P. Multimodality imaging of neurodegenerative disorders with a focus on multiparametric magnetic resonance and molecular imaging. \u003cem\u003eInsights into Imaging\u003c/em\u003e. \u003cb\u003e14\u003c/b\u003e (1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13244-022-01358-6\u003c/span\u003e\u003cspan address=\"10.1186/s13244-022-01358-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMostafa, A. M., Zakariah, M. \u0026amp; Aldakheel, E. A. Brain Tumor Segmentation Using Deep Learning on MRI Images. \u003cem\u003eDiagnostics\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e (9). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/diagnostics13091562\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13091562\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNobakht, S. et al. Combined atlas and convolutional neural network-based segmentation of the hippocampus from mri according to the adni harmonized protocol. \u003cem\u003eSensors\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e (7). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s21072427\u003c/span\u003e\u003cspan address=\"10.3390/s21072427\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRaji, C. A., Meysami, S., Porter, V. R., Merrill, D. A. \u0026amp; Mendez, M. F. Diagnostic utility of brain MRI volumetry in comparing traumatic brain injury, Alzheimer disease and behavioral variant frontotemporal dementia. \u003cem\u003eBMC Neurol.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e (1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12883-024-03844-4\u003c/span\u003e\u003cspan address=\"10.1186/s12883-024-03844-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRao, R., Paramasivam, G., Ramachandra Rao, I. \u0026amp; Prabhu, M. A. \u003cem\u003eNormality Testing in Statistics: What Clinician-Researchers Should Know\u003c/em\u003e. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/HFJI.HFJI_7_24\u003c/span\u003e\u003cspan address=\"10.4103/HFJI.HFJI_7_24\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSch\u0026ouml;ll, M. et al. Challenges in the practical implementation of blood biomarkers for Alzheimer\u0026rsquo;s disease. \u003cem\u003eLancet Healthy Longev.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.lanhl.2024.07.013\u003c/span\u003e\u003cspan address=\"10.1016/j.lanhl.2024.07.013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSghirripa, S. et al. Evaluating Hippocampal Segmentation Methods Evaluating Traditional, Deep Learning, and Subfield Methods for Automatically Segmenting the Hippocampus from MRI. Hum Brain Mapp. 1;46(5): e70200. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1101/2024.08.06.24311530\u003c/span\u003e\u003cspan address=\"10.1101/2024.08.06.24311530\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSong, J. et al. Development of Neurodegenerative Disease Diagnosis and Monitoring from Traditional to Digital Biomarkers. In \u003cem\u003eBiosensors\u003c/em\u003e (Vol. 15, Issue 2). Multidisciplinary Digital Publishing Institute (MDPI). (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/bios15020102\u003c/span\u003e\u003cspan address=\"10.3390/bios15020102\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSubramaniam, M. et al. Prevalence of dementia in Singapore: Changes across a decade. \u003cem\u003eAlzheimer\u0026rsquo;s Dement.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/alz.14485\u003c/span\u003e\u003cspan address=\"10.1002/alz.14485\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSui, C. et al. Decreased gray matter volume in the right middle temporal gyrus associated with cognitive dysfunction in preeclampsia superimposed on chronic hypertension. \u003cem\u003eFront. NeuroSci.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 1138952. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnins.2023.1138952\u003c/span\u003e\u003cspan address=\"10.3389/fnins.2023.1138952\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTavares, V., Prata, D. \u0026amp; Ferreira, H. A. Comparing SPM12 and CAT12 segmentation pipelines: a brain tissue volume-based age and Alzheimer\u0026rsquo;s disease study. \u003cem\u003eJ. Neurosci. Methods\u003c/em\u003e. \u003cb\u003e334\u003c/b\u003e, 108565. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.jneumeth.2019.108565\u003c/span\u003e\u003cspan address=\"10.1016/j.jneumeth.2019.108565\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThyreau, B., Sato, K., Fukuda, H. \u0026amp; Taki, Y. Segmentation of the hippocampus by transferring algorithmic knowledge for large cohort processing. \u003cem\u003eMed. Image. Anal.\u003c/em\u003e \u003cb\u003e43\u003c/b\u003e, 214\u0026ndash;228. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.media.2017.11.004\u003c/span\u003e\u003cspan address=\"10.1016/j.media.2017.11.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVejandla, B., Savani, S., Appalaneni, R., Veeravalli, R. S. \u0026amp; Gude, S. S. Alzheimer\u0026rsquo;s Disease: The Past, Present, and Future of a Globally Progressive Disease. \u003cem\u003eCureus\u003c/em\u003e. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7759/cureus.51705\u003c/span\u003e\u003cspan address=\"10.7759/cureus.51705\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXing, K., Ku, J. \u0026amp; Zhao, J. A Novel Approach to Optimizing Convolutional Neural Networks for Improved Digital Image Segmentation. \u003cem\u003eInt. J. Intell. Syst.\u003c/em\u003e \u003cb\u003e2024\u003c/b\u003e (1), 4337255. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1155/2024/4337255\u003c/span\u003e\u003cspan address=\"10.1155/2024/4337255\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhao, Z., Alzubaidi, L., Zhang, J., Duan, Y. \u0026amp; Gu, Y. A comparison review of transfer learning and self-supervised learning: Definitions, applications, advantages and limitations. \u003cem\u003eExpert Syst. Appl.\u003c/em\u003e \u003cb\u003e242\u003c/b\u003e, 122807. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.eswa.2023.122807\u003c/span\u003e\u003cspan address=\"10.1016/j.eswa.2023.122807\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6790322/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6790322/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe advent of artificial intelligence (AI) driven software has impacted numerous aspects of medicine, leading to automated algorithms that assist in performing feature extraction, making measurements on diagnostic imaging, and aiding in diagnosing disorders. AI-based convoluted neural networks (CNN) enable automated segmentation of the hippocampal volume seen on MRI diagnostic imaging, hence facilitating the diagnosis of Alzheimer\u0026rsquo;s disease (AD). Traditional voxel-based morphometry (VBM) used for measuring hippocampal volume can be time-laborious, thus CNN-based algorithms can minimize the time and reduce human errors. We utilized HippoDeep, an open-source CNN-based algorithm, to compare the hippocampal datasets from a Caucasian population with a dataset from a Southeast Asian AD and cognitively healthy control (HC) population. ROC analysis revealed superior diagnostic performance for HippoDeep, with AUCs of 0.918 (left hippocampus) and 0.882 (right hippocampus), compared to VBM\u0026rsquo;s 0.788 and 0.741, respectively. We determined cut-off thresholds for hippocampal volume to further improve the classification method. CNN-based method outperformed traditional semiautomated for segmentation accuracy (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) with insignificant interpopulation differences. Moreover, CNN-derived hippocampal volumes exhibited stronger correlations with MMSE scores (r\u0026thinsp;=\u0026thinsp;0.63 vs. r\u0026thinsp;=\u0026thinsp;0.42). HippoDeep offers accurate, reproducible, and generalizable hippocampal segmentation, supporting its potential as a clinical tool for early AD diagnosis across diverse populations.\u003c/p\u003e","manuscriptTitle":"Comparing Deep Learning CNN method with Traditional MRI-based Hippocampal segmentation and volumetry for Early Alzheimer’s Disease Diagnosis Across Diverse Populations","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-30 04:22:33","doi":"10.21203/rs.3.rs-6790322/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-08-25T10:11:07+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-23T05:21:24+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-12T20:34:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"324845900972348314489243024848639348384","date":"2025-08-12T20:14:26+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"171100872497447041275116962832021590057","date":"2025-08-08T01:31:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"297915713915353366783915560932220978171","date":"2025-08-04T19:49:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"148011396225150372498689360053924730939","date":"2025-08-04T13:38:45+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-25T07:51:44+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-17T08:48:01+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-10T03:27:53+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-06-10T03:25:09+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c4920bac-752b-4bc9-8b48-07172b71c071","owner":[],"postedDate":"July 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":52205517,"name":"Biological sciences/Neuroscience"},{"id":52205518,"name":"Biological sciences/Neuroscience/Cognitive ageing"},{"id":52205519,"name":"Health sciences/Diseases/Neurological disorders/Dementia"}],"tags":[],"updatedAt":"2025-12-08T16:00:41+00:00","versionOfRecord":{"articleIdentity":"rs-6790322","link":"https://doi.org/10.1038/s41598-025-29366-8","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-12-02 15:57:25","publishedOnDateReadable":"December 2nd, 2025"},"versionCreatedAt":"2025-07-30 04:22:33","video":"","vorDoi":"10.1038/s41598-025-29366-8","vorDoiUrl":"https://doi.org/10.1038/s41598-025-29366-8","workflowStages":[]},"version":"v1","identity":"rs-6790322","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6790322","identity":"rs-6790322","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}