Efficient Two Stage Segmentation Framework for Chest X-Ray Images With U-Net Model Fusion

Abstract

Medical imaging provides vital information about the respiratory system for health care diagnosis, counseling, and research purposes. Lung segmentation is essential for diagnosing lung conditions. Currently, deep learning algorithms are commonly used to segment images, assisting doctors in diagnosing lung diseases. However, this process is time-consuming, prone to errors, and requires medical expertise. Therefore, an enhanced two-stage segmentation framework for lung region segmentation is proposed in this research in order to improve the accuracy of lung area segmentation in chest X-rays. In the first phase, this framework utilizes deep learning models, specifically the U-Net and ResU-Net models, for the initial pre-segmentation of chest X-ray images. Segmented mask images obtained from the U-Net and ResU-Net models will undergo a Haar transform and spatial frequency fusion. Following image fusion, an inverse Haar transform will be applied, and typical image processing methods will be used for post-processing and refinement of the final segmented chest X-ray images. This approach aims to enhance the accuracy and quality of segmentation results, facilitating improved analysis and interpretation by medical professionals for diagnostic purposes. The proposed framework has demonstrated high accuracy rates of 98.42%, 98.28%, and 98.99% on JSRT, MC, and Shenzhen datasets, which are publicly available Chest X-Ray datasets.