Image segmentation of X-Ray and optical images using U-Net/UNet++ based deep learning architecture

Abstract

Image segmentation is a fundamental section of the current healthcare system to segment and detect diseases such as disease of the lung (pneumothorax), cancer, diabetic retinopathy, dengue, malaria, heart disease, Alzheimer’s disease, liver disease, rheumatoid arthritis, and so on. Lung segmentation, Cell segmentation, Brain segmentation, Liver segmentation are some of the popular medical segmentations. In this study, we worked on two different types of images x-ray image and optical image for lung (Pneumothorax) and cell (nucleus) image segmentation. For both cases, we employed U-Net++ for image classification and segmentation to detect and identify Pneumothorax or cell nuclei. Additionally, we incorporated several image recognition models U-Net, ResNet34, Inception V3 within U-Net++ architecture and investigated which model provides better accuracy with minimum loss. The findings of our study will be not only beneficial for clinicians for accurate diagnosis but also will be helpful to lessen diagnostic limitations.