Medical Image Segmentation Using Modified U-Net

Abstract

Medical image analysis is for essential in addressing medical puzzles by extracting respective image information from imaging devices to enhance diagnostic results. The current study proposes an Improved U-Net model named the Image Contrast-Based U-Net Segmentation model used to diagnose the skin diseases, as well as identifying the affected areas of the skin for segmentation. In the developed model, the red boundary box frames the areas affected to assist physicians in the identification and diagnosis of different skin diseases. Both of the used datasets ISIC and PH2 were used for training U-Net model and the test results of the model show that it achieves an accuracy of over 98% for identifying the affected pixels. The above system lets a user to upload test images, view original and segmented images, and the results of the same are clearly marked or highlighted on the areas to be affected. In the lamp-model performance comparison on the two sets of data we find equate performances. The underlying proposed system is an efficient, user friendly tool for skin disease identification which can be extended to other medical image segmentation applications coupled with efficiency increase to the diagnostic workflows in clinical practice.