A Multilayer Residual Dendritic Neural Model for Predicting Stroke Prognosis

Yuxi Wang, Haochang Jin, Maocheng Cao, Xiong Xiao, Li Wang

Abstract: Stroke, caused by occlusion or rupture of cerebral blood vessels, is a leading cause of disability and death globally. Accurate stroke prognosis can enhance clinical decisions and rehabilitation strategies. The dendritic neural model (DNM), inspired by biological neurons, shows strong predictive capability, but struggles with real-world small-scale tabular stroke data. Therefore, an improved residual dendritic neural model (RDNM) is proposed. It contains a series of stacked synaptic and dendritic layers to enhance the power. Residual connections are added between layers to address the vanishing gradient problem. Evaluations using one public and two private stroke prognosis datasets demonstrate that RDNM significantly outperforms original DNM and state-of-the-art deep-learning methods, highlighting its potential for clinical applications. Source code is available at https://github.com/jhc050998/RDNM.

External IDs:dblp:journals/access/WangJCXW25