Using artificial intelligence and radiomics to analyze imaging features of neurodegenerative diseases

TL;DR

This paper introduces an AI-based framework combining radiomics and symbolic reasoning to better analyze and predict the progression of neurodegenerative diseases like Alzheimer's and Parkinson's.

Contribution

The novel contribution is the NeuroSage framework, which integrates symbolic alignment and multi-modal data for improved disease modeling.

Findings

The framework outperformed existing methods with an F1 score of 88.90 on the ADNI dataset.

It achieved an F1 score of 85.43 on the PPMI dataset, showing effectiveness across multiple modalities.

The method demonstrates potential for real-world application in neurodegenerative disease monitoring.

Abstract

Neurodegenerative diseases such as Alzheimer's and Parkinson's are characterized by complex, multifactorial progression patterns that challenge early diagnosis and personalized treatment planning.

To address this, we propose an integrated AI-radiomics framework that combines symbolic reasoning, deep learning, and multi-modal feature alignment to model disease progression from structural imaging and behavioral data. The core of our method is a biologically informed architecture called NeuroSage, which incorporates radiomic features, clinical priors, and graph-based neural dynamics. We further introduce a symbolic alignment strategy (CAIS) to ensure clinical interpretability and cognitive coherence of the learned representations.

Experiments on multiple datasets—including ADNI, PPMI, and ABIDE for imaging, and YouTubePD and PDVD for behavioral signals—demonstrate that our approach consistently outperforms existing baselines, achieving an F1 score of 88.90 on ADNI and 85.43 on PPMI. These results highlight the framework's effectiveness in capturing disease patterns across imaging and non-imaging modalities, supporting its potential for real-world neurodegenerative disease monitoring and diagnosis.

Figures