摘要目的:基于深度学习算法和计算机断层扫描（CT）影像构建自发性脑出血（SICH）患者早期神经功能恶化（END）的智能预估系统。方法:回顾性分析，收集2022年1月至2024年10月于西安中医脑病医院进行治疗的369例SICH患者临床资料，入院后均行CT平扫，按照2∶1比例将患者随机分为试验集（246例，采用五折交叉验证分5个子集）和验证集（123例）。试验集男135例、女111例，年龄（56.79±8.82）岁；验证集男70例、女53例，年龄（57.85±9.12）岁。基于深度学习算法和试验集CT影像资料构建SICH患者END的智能预估系统，利用验证集分析该智能预估系统的诊断效能。结果:SICH患者END发生率为21.14%（78/369），其中试验集246例患者中END发生51例，验证集123例患者中END发生27例。试验集五折交叉验证的模型平均灵敏度、特异度、准确度分别为96.18%、97.44%、97.16%；经验证集测评，基于深度学习算法和CT影像构建SICH患者END的智能预估系统预测灵敏度、特异度及准确度分别为96.30%、97.92%、97.56%，且与END实际发生情况的一致性较高（Kappa=0.930， P<0.001）。 结论:基于深度学习算法和CT影像构建SICH患者END的智能预估系统对END具有较好的诊断效能，有一定的临床推广意义。

abstractsObjective:To construct an intelligent prediction system for early neurological function deterioration (END) in patients with spontaneous intracerebral hemorrhage (SICH) based on deep learning algorithms and computed tomography (CT) images.Methods:The clinical data of 369 patients with SICH who were treated in Xi 'an Encephalopathy Hospital of Traditional Chinese Medicine from January 2022 to October 2024 were retrospectively analyzed. All the patients underwent CT plain scan after admission. According to the ratio of 2:1, the patients were randomly divided into an experimental set (246 cases, 5 subsets by five-fold cross-validation) and a validation set (123 cases). There were 135 males and 111 females in the evperimental set, aged (56.79±8.82) years. There were 70 males and 53 females in the validation set, aged (57.85±9.12) years. The intelligent prediction system for END in the SICH patient was constructed based on deep learning algorithms and experimental CT image data, and the diagnostic performance of the intelligent prediction system was analyzed by the validation set.Results:The incidence rate of END in the SICH patients was 21.14% (78/369), with 51 cases occurring in the experimental group of 246 patients, and 27 cases occurring in the validation group of 123 patients. The average sensitivity, specificity, and accuracy of the model validated by 5-fold cross validation in the experimental set were 96.18%, 97.44%, and 97.16% respectively. According to the validation set evaluation, the diagnostic sensitivity, specificity, and accuracy of the intelligent prediction system for END in the SICH patient based on deep learning algorithms and CT images were 96.30%, 97.92%, and 97.56%, respectively, and the consistency with the actual occurrence of END was high (Kappa=0.930, P<0.001). Conclusion:The intelligent prediction system for END of SICH patients based on deep learning algorithms and CT images has good diagnostic performance for END, and it has a certain clinical promotion significance.