Personalized Treatment Outcome Assessment and Prognosis Prediction in Hemorrhagic Stroke: A Novel Approach Using Deep Learning Technology

Hemorrhagic stroke, ranking second in global mortality, remains a critical health concern. Despite a declining stroke incidence in China, the actual number of deaths and disability-adjusted life years lost has increased. Comprehending persistent disabilities in individuals who have survived a stroke is fundamental to the development of rehabilitation approaches, the economical allocation of resources for stroke treatment, and the evaluation of immediate stroke therapeutic interventions. The Modified Rankin Scale (mRS) at 90 days post-stroke is invaluable for outcome assessment. This study is dedicated to predicting 90-day mRS scores in hemorrhagic stroke patients. Currently, statistical methods dominate academic research for predicting 90-day mRS scores, but they entail extensive data collection, high computational costs, and sluggish response times. In contrast, advanced deep learning technology offers enhanced adaptability to diverse data types and models nonlinear relationships through complex neural networks. Firstly, a dataset based on pyradiomics data is constructed to assess the disability of stroke patients using the 90-day mRS, and applied two prediction models, generalised regression neural network (GRNN) and genetic algorithm optimised backpropagation neural network (GA-BP), to predict the 90-day mRS scores, which were validated using Spearman's coefficient. The goal is to comprehensively explore disease progression and treatment outcomes, contributing to accurate personalised outcome assessment and prognosis prediction. This research, while still pending clinical trials, holds substantial promise for future applications, potentially revolutionizing the prognostic landscape for patients with hemorrhagic stroke.