A Hybrid CNN Framework for Robust multiclass Alzheimer’s Disease Classification Using ADNI sMRI

Md.Shamim, Shahriar Chowdhury Imran, D.M Arif Afsar, MD. TANVIR AHMED SIDDIQUE

Keywords: Alzheimer’s Disease, Structural Magnetic Resonance Imaging, Hybrid Con volutional Neural Network, Explainable Artificial Intelligence, Grad-CAM++, Bias-Aware Learning, Neuroimaging Analysis.

TL;DR: A bias-aware, interpretable CNN detects early Alzheimer’s from MRI scans with ~98% accuracy, using Grad-CAM++ to highlight relevant brain regions.

Registration Requirement: Yes

Abstract: Alzheimer's disease (AD) is a progressive neurodegenerative disorder that severely affects memory, cognition, and daily functioning, making early detection crucial for effective intervention and disease management. This study proposes an interpretable and bias-aware deep learning framework for early Alzheimer's disease classification using structural magnetic resonance imaging (sMRI) data. A hybrid convolutional neural network (CNN) was developed to extract deep spatial features from pre-processed T1-weighted MRI scans, which included global statistical features including the mean and standard deviation of pixel intensities. The dataset was obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and underwent an extensive preprocessing pipeline, including N4 bias field correction, skull stripping, strict registration to MNI152 space, intensity normalization, and 2D slice extraction. To address class imbalance and fairness concerns, weighted sampling technique was used during training. The proposed model achieved high classification performance with a test accuracy of about 98\%, demonstrating strong generalization ability. Furthermore, Grad-Cam++ was applied to generate class-specific attention maps, which enhance interpretability by highlighting anatomically relevant brain regions associated with Alzheimer's pathology. The integration of interpretability and bias-aware techniques reduces the black-box limitations of deep learning models and promotes credible AI in medical diagnosis. Overall, this study contributes to an accurate, interpretable, and ethically responsible framework for early-stage Alzheimer's disease detection.

Reproducibility: https://github.com/MohammadShamim-29/Alzheimar-Disease-Classification-with-ADNI-Data/tree/main

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 122