CT-based hybrid deep learning–radiomics framework for predicting postoperative rebleeding in hypertensive intracerebral hemorrhage

Abstract

OBJECTIVES: Hypertensive intracerebral hemorrhage (HICH) is a frequently encountered and highly lethal cerebrovascular disorder, and postoperative rebleeding remains one of its most feared complications. METHODS: We developed a nomogram featuring a hybrid architecture that integrates radiomics-derived signatures and deep learning representations from CT imaging together with routine clinical parameters, with the goal of enhancing the individualized prediction of rebleeding risk following surgical intervention in HICH patients. A total of 151 individuals diagnosed with HICH were prospectively enrolled and randomly assigned to a training set (n=105) and a validation set (n=46) following a 7:3 ratio. RESULTS: The resulting model outperformed single-domain approaches relying solely on traditional clinical indicators or deep learning features. In the training cohort, the nomogram yielded an AUC of 0.993 (95 % CI: 0.982-1.000), while in the internal testing cohort, the AUC reached 0.860 (95 % CI: 0.745-0.974). The model highlighted several key predictors associated with postoperative rebleeding. CONCLUSIONS: Overall, the integrated nomogram, embedding clinical data, radiomic phenotypes, and deep learning markers, exhibited robust predictive capability in assessing rebleeding risk among patients with HICH. Ongoing research is needed to further refine and validate the model in broader clinical settings.

Ähnliche Arbeiten