smainfet / segmentation_lungs Goto Github PK

As part of the course Medical image processing, i developed a solution to perform an automatic segmentation of lungs affected by COVID-19.

Due to the scale of people affected by Covid-19 and because the scanners have great inter-variability, I proposed a robust and rapid lung segmentation method, using Deep Learning U-Net architecture, which will serve as a pre-processing step of a hypothetic chain of automatic diagnosis, to focus the automatic analysis of the CT scan images on the lung region.

comparison of segmentation results obtained with several methods

I used data available on kaggle labeled by expert in segmentation. To improve training, I removed the volumes where there were no lungs.

We can imagine that this processing chain will be used to direct health services as quickly and reliably as possible towards the diagnosis of the patient observed, and thus relieve them of this workload.