Downloads: 7 | Views: 155 | Weekly Hits: ⮙3 | Monthly Hits: ⮙3

Research Paper | Computer Science and Engineering | China | Volume 11 Issue 10, October 2022 | Rating: 6 / 10

Risk Assessment Network for Diabetic Cardiovascular Disease based on Causality and Individual Attention

Ming Zuo | Jingyi Deng | Liping Zhang | Qi Xu

Abstract: With the increasing number of diabetic patients worldwide, the prevention and treatment of diabetic cardiovascular disease, a major complication, has become a major social challenge. At present, most of the research on diabetic cardiovascular disease is based on statistical methods, focusing on the correlation analysis between the risk characteristics of patients, such as age and cholesterol, and the disease risk. This approach, which considers the individual characteristics of patients and the characteristics of metabolic indicators as the same risk characteristics, ignores the causal relationship between risk characteristics and disease risk, ignores the important information carried by the individual characteristics of patients and the background of diabetes, and further ignores the impact of differences in disease background. In order to fill this gap, we proposed a new deep learning model, namely, a risk assessment model for diabetic cardiovascular disease based on Causal stability and interaction of individual characteristics (causal-NET). The causally stable and time-aware Long short-term Memory network (Causal and time-aware TLSTM) was used to learn disease risk information in the metabolic characteristics of patients and enhance the stability of the model. Secondly, our model also designed an individual feature interaction layer, which used individual features to modify the disease information hidden information obtained by learning the Causal and time-aware TLSTM unit, so as to obtain a more accurate and comprehensive disease information representation for the risk assessment task of diabetic cardiovascular disease. Our experimental results demonstrate that the model presented here performs better in the diabetic CVD risk assessment task, and consistently outperforms the contrast model. The experimental evaluation indexes reached the model accuracy, recall, F1 score and 94.33%, 89.84%, 93.33% and 93.90% under the receiver operation feature curve, respectively.

Keywords: Diabetic cardiovascular disease, Metabolic characteristics selection, Individual characteristic interaction, Causal stable learning, Disease risk assessment

Edition: Volume 11 Issue 10, October 2022,

Pages: 184 - 195