3D Auto Segmentation Module for Ischemic Stroke Lesions from MONAI

Neuroimaging importance for stroke is growing widely. The difficulty of quantifying and describing ischemic stroke lesions is yet an unsolved and semi-automated time-consuming process. However, Ischemic Stroke Lesion Segmentation (ISLES) offered a program for researchers to analyze their solution for Ischemia lesion segmentation from 3D Magnetic Resonance Images (MRIs). This paper introduces a Deep Learning based automated system called 3D Auto Segmentation Module (3D ASM) for lesion identification and extraction, which utilizes two MRI imaging modalities named Diffusion Weighted Imaging (DWI) and Apparent Diffusion Coefficient (ADC) as input from ISLES 2022. It resampled all the images to the same resolution, then divided the data into seven random folds. Trained the data folds in the semantic segmentation network of SegResNet from MONAI and implemented the module along with Auto3DSeg. Five runs of seven-fold cross-validation results were obtained from the assemblage of twelve Convolution Neural Network models to make the final decision. The proposed 3D ASM achieved supremacy in the Dice Similarity Coefficient (DSC) of 0.854, among other existing segmentation techniques.