Explainable AD classification using integrated quantum-inspired deep neural and transformer models

A new study introduces an explainable and robust framework for classifying Alzheimer's disease (AD) using a combination of deep learning and quantum-inspired techniques. The proposed model, named Tri-Fusion-ADNet, integrates EfficientNet-V2-S for localized brain feature extraction, Swin TransformerV2 for modeling long-range contextual relationships, and a quantum-inspired variational neural network (QI-VNN) for enhanced higher-order feature interaction.

The researchers tested the model on four public brain MRI datasets containing various AD stages, including non-demented, mild cognitive impairment, and progressive Alzheimer's. Balanced and augmented scans were used, acquired with different imaging techniques and techniques, to account for heterogeneous neurodegenerative patterns and non-generalization issues across imaging sources.

The model underwent 10-fold cross-validation on multiple datasets to ensure reliability. Additionally, qualitative interpretability was achieved through Grad-CAM visualization. The results showed exceptional performance across all datasets, with testing accuracy ranging from 91.18% to 98.42%. Notably, when trained on one dataset and tested on another, the model maintained high accuracies of 91.22%, even with unseen-domain shifts, demonstrating its strong generalization capabilities.

This research, conducted at the Department of Computer Science and Engineering at Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology in India, was not funded by any specific grants. The findings were published under a Creative Commons Attribution 4.0 International License.