Recently, Professor Liu Yongguo's team from the School of Information and Software Engineering at the University of Electronic Science and Technology of China has made new progress in the auxiliary diagnosis of Alzheimer's disease, with the relevant findings published in the "IEEE Transactions on Image Processing."
Building an intelligent auxiliary diagnostic model based on brain imaging data can provide support for the early diagnosis and precise intervention of Alzheimer's disease, with disease stage identification and cognitive function prediction being two key issues in the auxiliary diagnosis of Alzheimer's disease.
The study proposes a joint feature selection model based on shared manifold regularization to identify Alzheimer's disease-related brain imaging features by exploiting the correlation between disease stage and cognitive function. For disease stage identification, a combination of linear discriminant analysis and subspace sparse regularization is proposed to identify disease stage-related brain imaging features.
Furthermore, the model introduces locally preserved intra-class scatter matrices and adaptive learning of local relationships between samples. For cognitive function prediction, a latent cognitive score space is established based on the correlation between cognitive scores, and a sparse regression model is trained with latent cognitive scores as targets to identify cognitive function-related brain imaging features. Additionally, global consistency and local consistency regularization terms are designed to guide the learning of the latent cognitive score space.
Since patients with similar cognitive function are often in the same disease stage, the model learns a shared graph structure for different tasks, fully exploiting the correlation between disease stage and cognitive function, thereby improving the efficiency of feature selection. Through simulation experiments on Alzheimer's disease brain imaging, researchers have verified the effectiveness of the proposed model.
Related paper information: https://doi.org/10.1109/TIP.2024.3382600
