Scholarlink

Objectives <p>To explore the value of a deep learning (DL) model based on whole-brain CT perfusion (CTP) maps at admission for predicting delayed cerebral ischemia (DCI) in patients with aneurysmal subarachnoid hemorrhage (aSAH).</p> Materials and methods <p>This study included 240 aSAH patients from XX Hospital who underwent whole-brain CTP at admission. The training and test sets were divided through stratified sampling based on the presence of DCI. Three models were developed: a clinical model, a DL-based imaging model from CTP maps, and a fusion model combining both. Two neuroradiologists also provided predictions using admission data. The performance of the models evaluated using the area under the receiver operating characteristic curve (AUC), with secondary metrics including sensitivity, specificity, and their 95% confidence intervals.</p> Results <p>Among the seven CTP map models, the FEP model demonstrated the highest AUC of 0.82 with a sensitivity of 49% and specificity of 86% in the validation set. Grad-CAM based on FEP maps indicated that the lateral ventricles and cortical areas in the periphery of the brain were the key regions for DL decision-making.</p> <p>In the test set, the fusion model outperformed both the clinical model (AUC, 0.82 vs. 0.73, <i>p</i> = 0.04) and neuroradiologists assessments (vs. Neuroradiologist A5: 0.82 vs. 0.61, <i>p</i> = 0.01; vs. Neuroradiologist A9: 0.82 vs. 0.63, <i>p</i> = 0.04) in predicting DCI, but showed comparable efficacy to the imaging model (AUC, 0.82 vs. 0.76, <i>p</i> = 0.42).</p> Conclusion <p>This study demonstrates that a DL-based model, trained on CTP data obtained within 24&#xa0;h of symptom onset, can effectively and automatically predict the development of DCI in aSAH patients, supporting clinical decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning based on CT perfusion maps predicts delayed cerebral ischemia at admission following aneurysmal subarachnoid hemorrhage

  • 期刊论文

  • 发表日期：

    2025年9月26日

查看全文
  • Qixing Wu,
  • Lei Feng,
  • Haifeng Cheng,
  • Fei Xiong,
  • Chao Zhang,
  • Fei Li,
  • Jinlong Yuan,
  • Mingquan Ye,
  • Yunfeng Zhou

摘要

Objectives

To explore the value of a deep learning (DL) model based on whole-brain CT perfusion (CTP) maps at admission for predicting delayed cerebral ischemia (DCI) in patients with aneurysmal subarachnoid hemorrhage (aSAH).

Materials and methods

This study included 240 aSAH patients from XX Hospital who underwent whole-brain CTP at admission. The training and test sets were divided through stratified sampling based on the presence of DCI. Three models were developed: a clinical model, a DL-based imaging model from CTP maps, and a fusion model combining both. Two neuroradiologists also provided predictions using admission data. The performance of the models evaluated using the area under the receiver operating characteristic curve (AUC), with secondary metrics including sensitivity, specificity, and their 95% confidence intervals.

Results

Among the seven CTP map models, the FEP model demonstrated the highest AUC of 0.82 with a sensitivity of 49% and specificity of 86% in the validation set. Grad-CAM based on FEP maps indicated that the lateral ventricles and cortical areas in the periphery of the brain were the key regions for DL decision-making.

In the test set, the fusion model outperformed both the clinical model (AUC, 0.82 vs. 0.73, p = 0.04) and neuroradiologists assessments (vs. Neuroradiologist A5: 0.82 vs. 0.61, p = 0.01; vs. Neuroradiologist A9: 0.82 vs. 0.63, p = 0.04) in predicting DCI, but showed comparable efficacy to the imaging model (AUC, 0.82 vs. 0.76, p = 0.42).

Conclusion

This study demonstrates that a DL-based model, trained on CTP data obtained within 24 h of symptom onset, can effectively and automatically predict the development of DCI in aSAH patients, supporting clinical decision-making.