Combination of Image Improvement on Segmentation Using a Convolutional Neural Network in Efforts to Detect Liver Disease

Abstract

Liver disease is a disease caused by various factors such as the spread of viruses. Liver damage causes the ability to break down red blood cells to be disrupted. Detection of liver disease can be done using the segmentation. Segmentation is useful for separating an area of the liver in an image from other areas. Segmentation carried out manually requires experts and a long time, so automatic segmentation is needed. CNN can be used to perform automatic segmentation. One of the CNN architectures is the U-Net architecture. Segmentation requires quality images to improve recognition of image patterns, so image improvement is needed in the form of contrast enhancement. Contrast improvement was carried out by taking Green Channel images. Contrast enhancement was carried out using the Contrast Stretching and CLAHE methods. The image improvement results show MSE and SSIM values 66.1844 and 0.7088. Evaluation of the image improvements obtained provides significant changes. The improved image is used at the segmentation stage. Segmentation is carried out using the U-Net architecture. The segmentation results obtained performance evaluation values in the form of accuracy 99.6%, sensitivity 98.9%, and specificity 99.7%. This shows that the proposed method can detect liver disease in liver images well