recogsal / ischemic-stroke-lesion-segmentation

Ischemic-Stroke-Lesion-Segmentation with Unet Families

This repository contains the implementation of Unet Families for the task of segmenting Ischemic Stroke Lesion.

Sample Output From Left to Right: Input Image, Ground Truth , Predicted Segmentation Mask

Repository Structure

• DLM_final_ipynb contains training , evaluation and experiment code

• layers.py contains some special layers implemented for networks. It need to be put in the same fold as DLM_final_ipynb

• cross-validation contains csv for five fold cross-validation

• train_dataloader_final_v3.csv and validate_dataloader_final_v3 are csv to be load for five channels

• train_dataloader_final.csv and validate_dataloader_final.csv are csv to be load for one channel

• All five-fold validation experiment and five channel experiment are run with same hyperparameters as one channel experiments

Methods

• Unet

• Attention Unet (2D/3D)

• Multi-resolution Unet (2D/3D)