AI Model Accurately Diagnoses Alzheimer's from Incomplete Brain Scans

Alzheimer's disease (AD) is a progressive neurodegenerative disorder requiring early and accurate diagnosis for effective management. Current diagnostic approaches often rely on multi-modality neuroimaging, combining various brain scan types like MRI and PET for comprehensive assessment. However, clinical practice frequently encounters incomplete datasets where one or more imaging modalities are missing, significantly hindering diagnostic accuracy and the utility of existing AI models. This study addresses the critical challenge of maintaining high diagnostic performance for Alzheimer's disease when faced with partial or incomplete multi-modality neuroimaging data.

The DSIP framework demonstrated superior diagnostic capabilities for Alzheimer's disease, particularly excelling in scenarios with incomplete data. For the critical AD vs. Normal Control (NC) classification, the model achieved an impressive accuracy of 94.2%, outperforming existing baseline models by up to 3.5%. In the more challenging distinction between AD and Mild Cognitive Impairment (MCI), the DSIP framework reached an accuracy of 89.1%, showing a significant improvement of 2.8% over the next best method. The DSIP framework consistently achieved higher diagnostic accuracy and robustness across various scenarios of missing data, demonstrating its unparalleled ability to effectively leverage partial information. Furthermore, the model exhibited a 2.1-fold increase in the preservation of domain-specific features, indicating its enhanced capacity to extract and utilize unique diagnostic insights from each imaging modality. These results underscore the potential for more reliable AD diagnosis in real-world clinical settings where complete imaging datasets are often unavailable.