An Integrated Genetic, Molecular, and Imaging Machine Learning Model for Predicting Neurorepair Potential and Guiding Personalized Rehabilitation After Intracerebral Hemorrhage: A Multicenter Retrospective Study

Abstract

Introduction: Intracerebral hemorrhage (ICH) is a devastating subtype of stroke characterized by high mortality and disability rates. Objective: The objective of this study is to develop and validate a machine learning (ML) predictive model integrating genetic polymorphism, serum neurotrophic factor levels, and neuroimaging features for predicting 3-month spontaneous neurological recovery potential which is defined as favorable recovery with modified Rankin Scale (mRS) ≤ 2 without intensive rehabilitation intervention in patients with ICH and to explore its value in guiding personalized rehabilitation strategies. Methods: A retrospective multicenter study was conducted in multiple tertiary hospitals. Baseline clinical data, brain-derived neurotrophic factor (BDNF) Val66Met gene polymorphism, serum BDNF and nerve growth factor (NGF) levels, and CT/MRI imaging features were collected. The mRS score at 3 months post-ICH was used as the primary outcome measure. Multiple ML algorithms, including logistic regression, random forest, and extreme gradient boosting (XGBoost), were applied to construct predictive models. Results: The XGBoost-based multidimensional model achieved the highest predictive efficacy, with an AUC of 0.892 (95% CI: 0.867–0.917) in the training cohort, 0.881 (95% CI: 0.848–0.914) in the internal validation cohort, and 0.875 (95% CI: 0.842–0.908) in the external validation cohort, significantly outperforming the conventional clinical model (AUC: 0.681, p < 0.001) and single-dimensional models (genetic model: AUC = 0.723; serum factor model: AUC = 0.756; imaging model: AUC = 0.782). DCA confirmed that the multidimensional model provided a higher net clinical benefit across a wide range of threshold probabilities. Stratification analysis revealed that patients with wild-type BDNF Val66Met genotype, high serum BDNF/NGF levels, and preserved white matter tract integrity had an 88.3% probability of favorable spontaneous recovery. Conclusion: The multidimensional predictive model developed in this study demonstrates excellent performance in predicting 3-month neurological outcomes after ICH, which can effectively stratify patients into spontaneous recovery and intensive rehabilitation-required subgroups, providing a reliable basis for personalized post-ICH rehabilitation management.