---

About

Interfaces

This algorithm implements all of the following input-output combinations:

Model Facts

Summary

DeepISLES is a deep learning-based algorithm for automated stroke lesion segmentation. It leverages an ensembling strategy, combining leading algorithms from the ISLES'22 challenge to enhance robustness and accuracy. The algorithm generates binary lesion masks and visualizations to support advanced stroke research.

Mechanism

Target population¶

Patients with suspected acute or sub-acute ischemic stroke, including MRI before (acute) or after (sub-acute) receiving reperfusion treatment.

Inputs¶

Note: DICOM inputs must be provided for each MR sequence as a single .zip file.

1. Diffusion Weighted MRI (DWI b=1000, 3D volume)
   (.nii, .nii.gz, .mha, zipped .dcm)

2. Apparent Diffusion Coefficient map (ADC, 3D volume)
   (.nii, .nii.gz, .mha, zipped .dcm)

3. Fluid-attenuated inversion recovery MRI (FLAIR, 3D volume)
   (.nii, .nii.gz, .mha, zipped .dcm)

4. Algorithm settings (deepisles_config.json)
   Set "skull_strip": True if images are raw. (default: False - assuming your volumes are already defaced)

Outputs¶

• Binary stroke lesion segmentation mask (.mha)
• Output screenshot (.png)

Validation and Performance

Performance results on the ISLES'22 test set.

DSC ↑	AVD (ml) ↓	F1 ↑	ALD ↓
0.82 ± 0.12	1.59 ± 4.40	0.86 ± 0.21	1.00 ± 3.00

DSC: Dice score; AVD: Absolute volume difference; F1: lesion-wise F1 score; ALD: absolute lesion count difference.

Uses and Directions

This algorithm is intended for research purposes only and is not approved for clinical use or medical diagnosis in healthcare centers.

For further information about the algorithm please refer to the preprint DeepISLES: A Clinically Validated Ischemic Stroke Segmentation Model from the ISLES’22 Challenge

Full code is available at https://github.com/ezequieldlrosa/DeepIsles

Warnings

Common Error Messages

Information on this algorithm has been provided by the Algorithm Editors, following the Model Facts labels guidelines from Sendak, M.P., Gao, M., Brajer, N. et al. Presenting machine learning model information to clinical end users with model facts labels. npj Digit. Med. 3, 41 (2020). 10.1038/s41746-020-0253-3