Indirect Volume Estimation for Acute Ischemic Stroke from Diffusion Weighted Image Using Slice Image Segmentation

Area of Science:

• Medical Imaging
• Neurology
• Artificial Intelligence in Medicine

Background:

• Accurate estimation of acute ischemic stroke (AIS) volume using diffusion-weighted imaging (DWI) is critical for patient assessment and treatment guidance.
• Existing methods may lack objectivity, speed, or accuracy in determining AIS lesion size from DWI.
• Developing automated tools can significantly improve the efficiency and reliability of AIS volume quantification.

Purpose of the Study:

• To propose and validate an objective, rapid, and accurate method for estimating AIS volume from DWI.
• To compare the performance of 2D (indirect) and 3D (direct) segmentation algorithms for AIS volume estimation.
• To assess the clinical utility of the developed model as a decision-support tool for physicians.

Main Methods:

• Development of algorithms using 3D segmentation (direct estimation) and 2D segmentation (indirect estimation) for AIS volume quantification.
• Utilized a large dataset of DWI scans from 2159 participants with various AIS types for internal and external validation.
• Compared algorithm performance against manual annotations by neurologists, evaluating segmentation metrics and volume estimation accuracy.

Main Results:

• The pretrained indirect model (2D segmentation) outperformed the direct model (3D segmentation) in segmentation performance across internal and external validation sets.
• The indirect model achieved high volume estimation reliability, with 93.3% volume similarity (VS) and 0.797 mean absolute error (MAE) in internal validation.
• External validation confirmed the indirect model's robustness, showing 89.2% VS and 2.5% MAE, demonstrating its potential for real-world application.

Conclusions:

• The indirect model utilizing 2D segmentation provides an accurate and reliable method for estimating AIS volume from DWI.
• This automated approach can serve as a valuable supporting tool for physicians in making critical clinical decisions for AIS patients.
• The study highlights the potential of AI-driven image analysis to enhance stroke care efficiency and accuracy.