VISUALIZING BIOMARKER SIGNATURES: EXPLAINABLE DEEP LEARNING FOR EARLY DETECTION OF ALZHEIMER’S AND NEURODEGENERATIVE DISORDERS

Read full text Summarise with VIRAT
Authors: Ravindra Keshav Moje, Sai Kiran Oruganti, S. Sudheer Khan, Santosh H. Lavate
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The timely and precise diagnosis of neurodegenerative diseases, especially Alzheimer disease and related disorders, is a major clinical issue because of the subtle changes in the pathology of the disease at the early stage, variable disease progression, and low interpretability of the current artificial intelligence based systems of diagnosis. To overcome these obstacles, this paper introduces XAI-BioNet, which is a more explainable deep learning network, allowing a combination of state-of-the-art feature selection, clear biomarkers discovery, and strict laboratory validation of early neurodegenerative disease detection. XAI-BioNet uses the multimodal data including structural MRI, longitudinal cognitive functions, and demographic characteristics. Complementary architectures, such as VGG and MobileNetV2 to learn deep features through hierarchical and efficient learning of spatial features on the basis of neuroimaging data and LSTM networks to learn temporal dynamics in cognitive progression are used to perform the deep feature extraction. The dimensionality of features and redundancy are covered by the principal component analysis and the recursive elimination of features that makes sure to retain biomarkers that are clinically relevant. SHAP and Grad-CAM allow obtaining model interpretability and making transparent attribution of the diagnostic decision and neuroanatomical relevance. The comprehensiveness of validation is carried out on heterogeneous, real world clinical data with stratified cross-validation and cross-cohort evaluation undertaken to determine the robustness, scalability and generalizability. The recommended XAI-BioNet framework has a 96.2% diagnostic accuracy, 94.7% sensitivity, 97.1% specificity, and AUC= 0.97, and it is more effective than standalone VGG (89.6% accuracy), MobileNetV2 (91.8% accuracy), and LSTM (92.9% accuracy) models. The findings emphasize the possibility of XAI-BioNet being a clinically understandable and scalable model of early diagnosis of neurodegenerative diseases.

Fields of Study & Subjects

Dementia and Cognitive Impairment Research Artificial Intelligence in Healthcare Explainable Artificial Intelligence (XAI)

Publication details

Journal / source
ShodhKosh Journal of Visual and Performing Arts
Volume
7
Issue
2s
Publication date
28 March 2026
Language
EN