UNet with Two-Fold Training for Effective Segmentation of Lung Section in Chest X-Ray

Abstract

Segmentation and evaluation of the Region of Interest (ROI) in medical imaging is a prime task for disease screening and decision-making. Due to accuracy, Convolutional-Neural-Network (CNN) based ROI segmentation has been widely employed in recent years to evaluate a class of medical images recorded using chosen modality. The proposed work aims to demonstrate the segmentation performance of the UNet scheme with a one-fold and two-fold training process. To experimentally verify the merit of the proposed scheme, segmentation of the lung section from the chest X-ray is studied. This research includes the following parts; (i) Resizing the test image and image mask to pixels, (ii) Training the UNet with one-fold and two-fold approaches, (iii) Extracting the ROI, (iv) Comparing the ROI with the mask to compute the image metrics and (v) Validating and confirming the segmentation performance of UNet. The performance of UNet is then verified with UNet+ and UNet++. The investigational ending substantiates that the proposed approach helps to get better Jaccard (>95%), Dice ((>97%), and Accuracy (>98%) in two-fold training compared to other methods considered in this study.