Integrated Residual with Combined Temporal Module U-Net for Alzheimer's Disease Progression Prediction

¹ Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology, Nitte (Deemed to be University), Bengaluru, India
² Department of Information Science and Engineering, Sri Krishna Institute of Technology, Bengaluru, India
³ Department of Electronics and Communication Engineering, Dayanand Sagar University, Bengaluru, India
⁴ Department of Computer Engineering and Applications, GLA University, Mathura, India
⁵ Department of Information Science and Engineering, Dayananda Sagar Academy of Technology and Management, Bengaluru, India

^* Corresponding author: namitha-ise@dsatm.edu.in

Abstract

In recent years, Alzheimer’s Disease (AD) is a serious brain condition that affects millions of people around the world, it is hard to diagnose and treat. But, recently, Deep Learning (DL) technique are showing promising in helping to predict disease progression. However, existing models struggle to find new types of patient data and medical tests haven’t seen before and complex to find spatial features, due to lack of visualization tools class mislabeling. Therefore, an Integrated Residual with Combined Temporal Module U-Net (IRCTMU-Net) used to solve this above problem. Then, the data is preprocessed and a Residual U-Net is used, consisting of an encoder and a decoder. The encoder is responsible for reducing the image size to extract important features, while decoder increases the size back to get the final segmentation map. A special module is placed between them to strengthen the extracted features. Next, the attention module. An attention module with attention gates is also added to capture both local and global relationships, helping the model learn more useful features. To test the proposed IRCTMU-Net, experiments were carried out on the ADNI dataset and compared with other existing models. Finally, proposed IRCTMU-Net achieved better results in terms of accuracy (99.80%) and precision (99.83%) respectively when compared with exiting model like Temporal Graph Attention (TGN).