Lung chest x-ray image segmentation for detection of pneumonia using convolutional neural network

Abstract

Pneumonia has been identified as the top cause of mortality in children under the age of five, as well as in elderly with comorbidities. According to the World Health Organization, pneumonia reported 14% fatalities in children under the age of five nationwide in 2019. Chest x-ray (CXR) has been commonly used for detection of pneumonia. However, factor such as noise with low levels of intensity and low contrast between the images and the boundary representation can modify CXR images and it also requires highly skilled medical practitioners to accurately interpret the CXR images. Therefore, the goal of this study is to develop an automatic segmentation model to segment the region of interest (ROI) of pneumonia lung CXR images using U-Net architecture. Image enhancement using Contrast Limited Adaptive Histogram Equalisation (CLAHE) and gamma-correction based enhancement technique were applied to increase the quality of CXR images. Statistical analysis on features extracted from the segmented lung CXR images was performed to analyze the performance of the model was developed. The U-Net segmentation model achieves 95.58%, 95.82% and 95.48% accuracy for normal CXR while the model achieves 86.76%, 87.98% and 86.21% accuracy for pneumonia CXR which indicate that the U-Net segmentation for CLAHE x-ray images has better performance in segmenting the ROI of the lungs. As a conclusion, the segmentation model proposed shown to be able to overcome the disadvantages of manual segmentation where the model can be used to perform segmentation automatically on many CXRs at a time.