Stroke, a leading cause of death and disability globally, disrupts blood flow to a portion of the brain. This project explores recent advancements in deep learning for automated stroke lesion segmentation, focusing on four novel architectures: X-Net, V-Net, SegNet, and U-Net. The effectiveness of these models is evaluated on the Anatomical Tracings of Lesions After Stroke (ATLAS) dataset.
Stroke is a devastating neurological event, necessitating accurate analysis of stroke lesions in brain scans for effective treatment planning and patient outcomes. Deep learning offers promising tools for automated stroke lesion segmentation, providing advantages over traditional manual methods. This project delves into the potential of deep learning to automate stroke lesion segmentation using X-Net, V-Net, SegNet, and U-Net architectures.
- Evaluate the effectiveness of X-Net, V-Net, SegNet, and U-Net in accurately segmenting stroke lesions.
- Compare the performance of these models on the ATLAS dataset using appropriate evaluation metrics.
- Identify strengths and limitations of each model for stroke lesion segmentation in clinical settings.
- Utilized the ATLAS dataset for training and evaluation.
- Implemented preprocessing, model training, and evaluation pipelines.
- Evaluated models' performance using various segmentation metrics.
- Explored machine learning techniques for stroke prediction.
- Reviewed deep learning architectures like U-Net, V-Net, X-Net and SegNet for medical image segmentation.
- Highlighted key contributions of X-Net in medical image processing.
- Described data preprocessing steps and model training processes.
- Outlined the implementation of X-Net, U-Net, V-Net, and Seg-Net models.
- Clone the repository.
- Execute the respective scripts for model training and evaluation.
This project is licensed under the MIT License.