Semantic Segmentation of the Lung to Examine the Effect of COVID-19 Using UNET Model

Abstract

The Covid-19 pandemic is a universal problem that has caused significant outbreaks in every country and region, affecting men and women of all ages around the world. The automatic detection of lung infection is a major challenge that poses a limitation to the potential medical imaging offers to augment patient treatments and strategies for tackling the impact of Covid-19. One of the best and fastest way to diagnose this virus on a patient is to detect it on lung computed tomography (CT) scan images. Although, to find the tissues that are infected and segmenting them on the CT scan images face many challenges. To overcome these challenges, a method was created to enhance the slides on the CT scans, then a region of interest in which the lung was cropped out of the CT images to reduce the noise of the dataset before fitting it into the model for training. Due to the small amount of data, a method was utilized for data augmentation to overcome the problem of overfitting. After compiling the Unet model on the dataset and evaluating the model metrics, the results and the output that was generated show that the model achieved good results. The model achieved an Accuracy of approximately 96%, Intersection over Union (IoU) of approximately 85%, Dice Similarity Coefficient of approximately 92%, Precision of 92%, Recall of 93%, F1 score of 93%, and Loss of -85%.