Explainable AI Driven Multimodal Framework for Robust Spinal Muscular Atrophy Detection

Abstract: Spinal Muscular Atrophy is a serious hereditary disease which is a result of the mutations in the Survival Motor Neuron 1 gene (denotes the primary gene of the disease) which causes the gradual degeneration of motor neurons. Early diagnosis and prompt diagnosis is crucial to enhance patient outcomes and to facilitate timely medical care. Nevertheless, the conventional diagnostic techniques used to diagnose Spinal Muscular Atrophy are costly, lengthy and inaccessible and thus lead to late diagnosis. Moreover, there is a lack of effective integration of various clinical sources of data using current methods, and this reduces diagnostic accuracy and reliability.
To resolve these issues, this paper introduces a new multimodal deep learning system named Multimodal Attention-based Fusion Network, combining genetic information, medical images, electrophysiological measurements and clinical annotations. The model also includes attention mechanisms and the Explainable Artificial Intelligence based on SHAP to improve the interpretability and performance. The suggested framework was adopted and used by integrating various data modalities, such as survival motor neuron gene copy numbers, magnetic resonance images and ultrasound images, electrical impedance myography signals, and patient clinical history like age, functional score, and family history. The accuracy of 95.8 percent and the recall of 96.1 percent were demonstrated in an experimental evaluation, that was higher than traditional and single-modality methods, and the study was conducted in accordance with PRISMA guidelines, which guaranteed a systematic and validated research methodology.

Keywords: Spinal Muscular Atrophy, Multimodal AI, Explainable AI, MAF-Net, Deep Learning, Systematic Review