CHEST X-RAY IMAGE SEGMENTATION USING MODIFIED DEEPLABV3+ METHOD

Abstract

COVID-19 is a disease that affects the human respiratory system. The latest developments in September 2022 the number of confirmed cases of COVID-19 worldwide reached 608,328,548 with 6,501,469 patients who died. While in Indonesia confirmed COVID-19 reached 6,408,806 with 157,892 patients who died. Reserve Transcription Polymerase Chain Reaction (RT-PCR) is the most widely used tool. However, the latest RT-PCR test report shows that the RT-PCR test is inadequate. As an alternative, radiographic images such as chest x-rays and CT scans can help detect this. Radiographic images, especially x-rays, need processing to be able to make a diagnosis. Computer Aided Diagnosis (CAD) is a computer assisted diagnosis system that can be used as supporting information in making a diagnosis. To make it easier to make a diagnosis, we need a deep learning model that can help with this. DeepLabV3+ is a method that can carry out the segmentation process. DeepLabV3+ which is an extension of DeepLabV3 with the aim of improving the segmentation results. DeepLabV3+ uses a modified Xception as the backbone. In this study, 1,500 chest x-ray image data were used which were then divided into 80% for training data and 20% for testing data. There are 4 test scenarios in this study, namely with a learning rate of 0.01 without CLAHE, a learning rate of 0,01 and using CLAHE, a learning rate of 0,0001 without CLAHE, and a learning rate of 0,0001 using CLAHE. Of the 4 scenarios the learning rate scenario is 0,01 and using CLAHE gets the highest evaluation results using the Dice Similarity Coefficient (DSC) of 96.91%.