Lung CT Image Segmentation with Optimal Parameters Using UNET Architecture

In the modern era of medical science, Computed Tomography (CT) image segmentation is an important part that helps to extract the relevant information and supports for meaningful diagnosis. It is important to verify the intensity differences within a CT scan image and determine the region of interest efficiently. With the involvement and improvement of artificial intelligence, deep neural network based approaches have gained the attraction of researchers in the domain of computer-aided medical treatment. Artificial neural network, convolutional neural network are few examples of them. UNET architecture is one of the most widely used deep neural network model, that is successfully used in various image segmentation based diagnosis and treatments. Manual optimization of CT image segmentation is very time-consuming. In order to avoid such manual optimization, in this research work, we use UNET architecture for automatic CT image (specifically, lungs' CT images) segmentation. We apply that architecture on a set of exclusive CT image data and verify the framework that will help the experts to distinguish and early diagnosis of diseases, viz., lung cancer etc. We generate the predicted CT images using UNET and subsequently validate each prediction against the generated masked images of each exclusive CT image.