Ishemic Stroke Lesion Segmentation by Analyzing MRI Images Using Dilated and Transposed Convolutions in Convolutional Neural Networks

Abstract

Convolution Neural Networks(CNN/ConvNet) have raised the standard of semantic segmentation and is approaching human accuracy. Improvements in CNN architectures have led to wide spread adaptation of CNN in solving medical imaging problems. Automatic detection of various tumors through MRI Imaging techniques such as MRI and CT scans have very large scale impacts on healthcare sector. An encoder-decoder CNN with dilated convolution for ishemic lesion segmentation task is implemented and evaluated. The dice score and jacard score achieved are 0.85 ad 0.78 respectively.