Spatio-Temporal Deep Learning for Final Infarct Prediction using Acute Stroke CT Perfusion Data

TensorFlow implementation of our method for the ISLES 2024 Challenge: "Spatio-Temporal Deep Learning for Final Infarct Prediction using Acute Stroke CT Perfusion Data."

Abstract

Accurate prediction of the tissue outcome is crucial for guiding treatment decisions in acute ischemic stroke (AIS). Spatio-temporal (4D) Computed Tomography Perfusion (CTP) provides detailed insights into cerebral blood flow dynamics, which are essential for predicting final infarct regions. However, its high-dimensional and noisy nature presents challenges for direct prediction. In this study, we evaluate a deep learning model that fully leverages 4D CTP data for predicting tissue outcomes. The model integrates a shared-weight convolutional neural network (CNN) encoder, a Transformer encoder, and a CNN decoder to capture both spatial and temporal dependencies within the data. We evaluated this approach on a multicenter dataset of 143 patients from the ISLES 2024 challenge. The results reveal a Dice score of 0.20, an absolute volume difference of 17 ml, a lesion count difference of 19, and a lesion-wise F1-Score of 0.02, underscoring both the potential and challenges of directly utilizing 4D CTP data for final infarct prediction.

Usage

Installation

Recommended environment:

• Python 3.8.1
• TensorFlow GPU 2.4.1
• CUDA 11.0.2
• cuDNN 8.0.4.30

To install the dependencies, run:

$ git clone https://github.com/kimberly-amador/ISLES24-PrediCTP
$ cd ISLES24-PrediCTP
$ pip install -r requirements.txt

Data Preparation

1. Download the ISLES24 dataset from the official challenge page on the Grand Challenge website.
2. Preprocess the dataset using the included preprocessing script.

Model Training

1. Modify the model configuration. The default configuration parameters are in ./model/config.py.
2. Run python ./model/main_unimodal.py to train the model.

Inference and Evaluation

1. Convert the model output back to images using the included postprocessing script.
2. Visualize the images using your favorite image viewer, or calculate the metrics used in the ISLES24 challenge using the included evaluation script.

License

This repository is released under the Apache 2.0 license as found in the LICENSE file.

Acknowledgement

Part of the code is adapted from open-source codebase:

• Transformer: https://github.com/keras-team/keras-io/blob/master/examples/vision/video_transformers.py
• Dice Loss: https://github.com/voxelmorph/voxelmorph/blob/legacy/ext/neuron/neuron/metrics.py
• Evaluation Metrics: https://github.com/ezequieldlrosa/isles24/tree/main