ISLES 2016 and 2017-Benchmarking Ischemic Stroke Lesion Outcome Prediction Based on Multispectral MRI.

Area of Science:

• Medical imaging analysis
• Neurology
• Machine learning in healthcare

Background:

• Comparing new and existing models for stroke lesion prediction is difficult due to variations in datasets and algorithms.
• Establishing a fair benchmark is crucial for advancing research in ischemic stroke outcome prediction.

Purpose of the Study:

• To address the challenge of model comparability in ischemic stroke lesion segmentation.
• To provide a standardized platform for evaluating and comparing lesion outcome prediction algorithms.

Main Methods:

• The Ischemic Stroke Lesion Segmentation (ISLES) challenge provided uniformly pre-processed datasets for lesion outcome prediction.
• Participating teams applied their algorithms to the ISLES datasets for evaluation.
• Performance was assessed through a fair and transparent evaluation system.

Main Results:

• Deep learning tools, primarily convolutional neural networks (CNNs), were employed by top-ranked teams.
• The ISLES challenge facilitated the identification of state-of-the-art approaches in lesion outcome prediction.
• Despite advancements, accurately predicting stroke lesion outcomes remains a complex task.

Conclusions:

• The ISLES challenge successfully established a benchmark for comparing ischemic stroke lesion segmentation models.
• Publicly available datasets and the online evaluation system continue to serve as a valuable resource for researchers.
• Ongoing efforts and advanced methods like deep learning are necessary to overcome the challenges in lesion outcome prediction.