High performance segmentation in chest radiographs using convolutional neural network and investigation of its effect on pneumonia detection

Pneumonia, also known as pneumonia, is a life-threatening lung disease caused by a bacterial or viral infection. Therefore, early diagnosis of the disease significantly reduces the mortality rate. Chest X-rays are of great importance in the diagnosis of the disease. Apart from the studies in the field of health in the diagnosis of pneumonia, many studies are carried out in the field of engineering and provide convenience in accelerating the diagnosis process. One of these applications is to diagnose the disease with Deep Learning methods. Tensorflow Library on Python programming language and open access datasets were used in the study. Dataset-1, which contains chest x-rays of children aged 1-5, and Dataset – 2, which also contains the marked data of some images from this dataset, were used as dataset. A two-stage Convolutional Neural Network model was created. In the first stage, a model was created for dataset-1 images to be segmented using modified U-Net architecture on Dataset – 2 with marked lung region. Segmentation was done by using the created model in Dataset – 1. In addition, an approach that can detect the thorax region is proposed int the last part of study. Segmentation accuracy was calculated as 98.98%. The designed Convolutional Neural Network model was applied to the dataset-2, and the test accuracy in detecting pneumonia was 98.55%.