Objective Estimating early lesion progression in ischemic stroke is essential for assessing thrombolytic treatment efficacy. While computed tomography perfusion (CTP) and diffusion-weighted imaging (DWI) are commonly used to determine irreversible tissue injury (core), their high costs and limited availability present challenges. An alternative is quantitative net water uptake (NWU) derived from non-contrast-enhanced CT (NCCT), a validated biomarker for predicting patient outcomes. However, current NWU measurements require time-consuming manual postprocessing, and automating this process remains challenging. Material and methods This retrospective study introduces a fully automated method for NWU quantification using only NCCT images. The proposed image processing pipeline is based on expert-defined heuristics and voxel-wise NWU calculations, without employing deep learning-based components. The method was developed and validated on an in-house dataset of 185 patients (155 after exclusions) and externally tested on 51 patients (46 after exclusions). Performance was assessed by comparing computed lesion masks with expert-annotated core lesion masks from CTP (in-house) and DWI (external), using lesion detection rate, mean absolute NWU error, Dice similarity coefficient, and mean average precision. Results Results demonstrated a high lesion detection rate of 94% (in-house) and 100% (external). The mean absolute NWU estimation error was 1.0% for in-house data and 1.5% for external data. Segmentation accuracy was moderate, with Dice similarity coefficients of 0.48 (in-house) and 0.47 (external). Conclusion This method provides a practical and interpretable solution for NWU quantification, allowing stroke assessment directly from NCCT without reliance on additional imaging modalities. Key Points <Emphasis Type="BoldItalic">Question</Emphasis> Current approaches for quantifying early lesion progression in ischemic stroke rely on manual imaging-based NWU estimation, underscoring the need for an automated and reproducible method. <Emphasis Type="BoldItalic">Findings</Emphasis> The method achieved high lesion detection accuracy and reliable net water uptake estimation, demonstrating its potential as a tool for automated stroke assessment. <Emphasis Type="BoldItalic">Clinical relevance</Emphasis> This study presents an automated NCCT-based method for stroke assessment that enables net water uptake estimation directly from routinely acquired scans. The approach provides an interpretable alternative to methods requiring multimodal imaging, supporting broader clinical applicability. Graphical Abstract

Fully automated quantification of net water uptake in acute ischemic stroke using only non-contrast CT imaging

期刊论文
出版模式：

开放获取
发表日期：

2025年12月25日

查看全文

Thilo Sentker,
Maximilian Nielsen,
Susan Klapproth,
André Kemmling,
Michael H. Lev,
Gabriel Broocks,
René Werner

摘要

Objective

Estimating early lesion progression in ischemic stroke is essential for assessing thrombolytic treatment efficacy. While computed tomography perfusion (CTP) and diffusion-weighted imaging (DWI) are commonly used to determine irreversible tissue injury (core), their high costs and limited availability present challenges. An alternative is quantitative net water uptake (NWU) derived from non-contrast-enhanced CT (NCCT), a validated biomarker for predicting patient outcomes. However, current NWU measurements require time-consuming manual postprocessing, and automating this process remains challenging.

Material and methods

This retrospective study introduces a fully automated method for NWU quantification using only NCCT images. The proposed image processing pipeline is based on expert-defined heuristics and voxel-wise NWU calculations, without employing deep learning-based components. The method was developed and validated on an in-house dataset of 185 patients (155 after exclusions) and externally tested on 51 patients (46 after exclusions). Performance was assessed by comparing computed lesion masks with expert-annotated core lesion masks from CTP (in-house) and DWI (external), using lesion detection rate, mean absolute NWU error, Dice similarity coefficient, and mean average precision.

Results

Results demonstrated a high lesion detection rate of 94% (in-house) and 100% (external). The mean absolute NWU estimation error was 1.0% for in-house data and 1.5% for external data. Segmentation accuracy was moderate, with Dice similarity coefficients of 0.48 (in-house) and 0.47 (external).

Conclusion

This method provides a practical and interpretable solution for NWU quantification, allowing stroke assessment directly from NCCT without reliance on additional imaging modalities.

Key Points

Question Current approaches for quantifying early lesion progression in ischemic stroke rely on manual imaging-based NWU estimation, underscoring the need for an automated and reproducible method.

Findings The method achieved high lesion detection accuracy and reliable net water uptake estimation, demonstrating its potential as a tool for automated stroke assessment.

Clinical relevance This study presents an automated NCCT-based method for stroke assessment that enables net water uptake estimation directly from routinely acquired scans. The approach provides an interpretable alternative to methods requiring multimodal imaging, supporting broader clinical applicability.

Graphical Abstract