Collaborative multitasking framework for enhanced hippocampus segmentation and Alzheimer's disease classification.

The early diagnosis of Alzheimer's disease has faced significant challenges, as the initial patients have hidden symptoms that are difficult to distinguish from conventional symptoms. In view of this, this article designs a collaborative multitasking algorithm framework that implements a positive feedback loop between classification tasks, significantly improving processing accuracy. Specifically, the algorithm consists of three sub networks: the initial segmentation sub network accurately identifies the hippocampus boundary and generates the initial segmentation mask; The classification subnetwork relies on initial segmentation information to effectively distinguish different stages of Alzheimer's disease; Finally, the fine segmentation sub network finely adjusts the contour of the hippocampus based on the classification results. To verify the superiority of this method, this study used 269 MRI sample of Alzheimer's disease patients, including clinical and public datasets. The experimental results demonstrate that the proposed method exhibits superior performance in both hippocampal classification and segmentation tasks. Specifically, in terms of segmentation, the method achieved an average Dice Similarity Coefficient (DSC) of 94.0% and a Jaccard Index (JA) of 80.6%. For classification tasks, the method demonstrated an accuracy (AC) of 98.8%, sensitivity (SEN) of 98.8%, specificity (SP) of 98.6%, and F1 score (F1) of 97.8%, collectively indicating excellent clinical performance.