ISLES (Ischemic Stroke Lesion Segmentation/Prediction) Challenge Datasets (2015, 2016, 2017, 2018)

ISLES (Ischemic Stroke Lesion Segmentation/Prediction) Challenge Datasets (2015, 2016, 2017, 2018)

The ISLES (Ischemic Stroke Lesion Segmentation/Prediction) Challenge is a continuous, globally recognized initiative in the medical image computing community. Launched in 2015, ISLES provides standardized, high-quality, multi-center datasets to serve as public benchmarks for the development and objective comparison of novel algorithms.

Specifically, the challenge focuses on automating the complex tasks of segmenting ischemic stroke lesions and predicting patient outcomes using multi-parametric Magnetic Resonance Imaging (MRI). By ensuring a level playing field, ISLES has played a critical role in advancing translational AI methods for clinical stroke analysis.

1. Overview and Rationale

This dataset is a compilation of the official training and testing data from the Ischemic Stroke Lesion (ISLES) Challenge series, spanning the years 2015, 2016, 2017, and 2018. The ISLES challenges were organized in conjunction with major medical imaging conferences, such as MICCAI, to establish public benchmarks for developing and evaluating algorithms in automated ischemic stroke analysis from multispectral Magnetic Resonance Imaging (MRI).

The primary focus areas include:

• ISLES 2015: Segmentation of sub-acute lesions (SISS) and estimation of acute perfusion lesions (SPES).

• ISLES 2016, 2017, 2018: Prediction of stroke lesion outcome based on acute multispectral MRI.

The data consists of anonymized, multi-center, multi-parametric MRI scans, with corresponding expert-annotated ground truth masks in the training sets.

2. Data History and Format

Previous Repository: This data was originally hosted on the SICAS Medical Image Repository (SMIR)platform (smir.ch), which is no longer active. This Zenodo release ensures the continued availability and accessibility of these crucial benchmark datasets for the scientific community.

Challenge Website: For detailed competition rules, evaluation metrics, and leaderboard results, please refer to the official challenge website: https://www.isles-challenge.org/

File Format: All imaging data is provided in NIfTI (.nii) format.

The root folders correspond to the challenge years: 2015, 2016, 2017, and 2018.

ISLES 2015

The 2015 challenge focused on two sub-challenges: Sub-Acute Stroke Lesion Segmentation (SISS) and Stroke Perfusion Estimation (SPES).

Internal Structure (Example: SISS):

Inside each testing folder, cases are organized by case number (<case number>). Within the case folder, you will find four sub-folders, each containing a .nii volume for a specific modality.

Note: The trailing number in the folder name (<ID>) is an internal system ID and is not relevant for data interpretation.

ISLES 2016

The 2016 challenge shifted focus to Lesion Outcome Prediction based on multi-parametric MRI from the acute phase. The folders distinguish between training/testing sets and between native and co-registered spatial domains.

Registration Type:

• Native: The MRI data is in the patient's original, acquired space.

• CoRegistered: The MRI data has been registered to a common anatomical reference or template, providing spatial alignment across modalities and patients.

Modalities Included (Perfusion and Diffusion Maps):

The dataset includes the following perfusion and diffusion maps: PWI, ADC, MTT (Mean Transit Time), rCBF (relative Cerebral Blood Flow), rCBV (relative Cerebral Blood Volume), TMax (Time-to-Maximum), and TTP (Time-to-Peak).

ISLES 2017 and ISLES 2018

The 2017 and 2018 challenges continued the focus on lesion outcome prediction and segmentation tasks. The data is organized into simple Training and Testing splits.

4. Latest Research and Recommended Model: DeepISLES

The ISLES challenges have driven the development of state-of-the-art AI algorithms. We encourage users of this dataset to explore the DeepISLES model, a robust, clinically validated solution derived from top submissions to the ISLES'22 challenge.

DeepISLES is a high-performing ensemble algorithm that achieves superior accuracy and generalizability in detecting and segmenting ischemic lesions from DWI and other MRI sequences. In external validation, neuroradiologists even showed a preference for the algorithm's segmentations over manual expert efforts in a Turing-like test, highlighting its potential for real-world clinical utility.

For full details on the model, its validation, and to access the software, please refer to our recent publication:

Publication:

DeepISLES: a clinically validated ischemic stroke segmentation model from the ISLES’22 challenge

Nature Communications

https://www.nature.com/articles/s41467-025-62373-x

5. Recommended Citations

Please cite the relevant ISLES challenge paper(s) when using this dataset in your work:

• For ISLES 2015 Data:

  Maier, O., Menze, B. H., von der Gablentz, J., et al. (2017). ISLES 2015 - A public evaluation benchmark for ischemic stroke lesion segmentation from multispectral MRI. Medical Image Analysis, 35, 250–269.

• For ISLES 2016 and 2017 Data:

  Winzeck, S., Hakim, A., Acland, P., et al. (2018). ISLES 2016 & 2017 - Benchmarking Ischemic Stroke Lesion Outcome Prediction Based on Multispectral MRI. Frontiers in Neurology, 9(679).

• For DeepISLES Model or related methods:

  De la Rosa, E., Reyes, M., Liew, S. L., et al. (2025). DeepISLES: a clinically validated ischemic stroke segmentation model from the ISLES’22 challenge. Nature Communications, 16(1), 7357.

Keywords

Ischemic Stroke, MRI, Multispectral Imaging, Segmentation, Lesion Outcome Prediction, ISLES Challenge, NIfTI, Medical Imaging, Deep Learning, Benchmarking, DWI, FLAIR, Perfusion, SMIR.