Scholarlink

Accurate prediction of the tissue outcome is crucial for guiding treatment decisions in acute ischemic stroke (AIS). Spatio-temporal (4D) Computed Tomography Perfusion (CTP) provides detailed insights into cerebral blood flow dynamics, which are essential for predicting final infarct regions. However, its high-dimensional and noisy nature presents challenges for direct prediction. In this study, we evaluate a deep learning model that fully leverages 4D CTP data for predicting tissue outcomes. The model integrates a shared-weight convolutional neural network (CNN) encoder, a Transformer encoder, and a CNN decoder to capture both spatial and temporal dependencies within the data. We evaluated this approach on a multicenter dataset of 143 patients from the ISLES 2024 challenge. The results reveal a Dice score of 0.20, an absolute volume difference of 17 ml, a mean lesion count difference of 19, and a lesion-wise F1-Score of 0.02, underscoring both the potential and challenges of directly utilizing 4D CTP data for final infarct prediction.

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

Spatio-Temporal Deep Learning for Final Infarct Prediction Using Acute Stroke CT Perfusion Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kimberly Amador,
  • Anthony J. Winder,
  • Nils D. Forkert

摘要

Accurate prediction of the tissue outcome is crucial for guiding treatment decisions in acute ischemic stroke (AIS). Spatio-temporal (4D) Computed Tomography Perfusion (CTP) provides detailed insights into cerebral blood flow dynamics, which are essential for predicting final infarct regions. However, its high-dimensional and noisy nature presents challenges for direct prediction. In this study, we evaluate a deep learning model that fully leverages 4D CTP data for predicting tissue outcomes. The model integrates a shared-weight convolutional neural network (CNN) encoder, a Transformer encoder, and a CNN decoder to capture both spatial and temporal dependencies within the data. We evaluated this approach on a multicenter dataset of 143 patients from the ISLES 2024 challenge. The results reveal a Dice score of 0.20, an absolute volume difference of 17 ml, a mean lesion count difference of 19, and a lesion-wise F1-Score of 0.02, underscoring both the potential and challenges of directly utilizing 4D CTP data for final infarct prediction.