Machine Learning Fusion-Based Detection Techniques for categorizing Abnormalities in Biomedical Images of the Lungs

Abstract

Thanks to technological advancements, assistive healthcare systems are mushrooming and assisting medical practitioners. In the last ten years, there has been a lot of buzz about the prospect of using AI and closely related technologies for proactive illness detection. By examining X-rays of the lungs, doctors can often identify tuberculosis (TB). Classification utilizing deep learning algorithms efficiently achieves TB detection accuracy comparable to that of a doctor. Classification algorithms used to segmented lungs rather than the full X-ray improve the likelihood of TB detection. The originality of this work is in the comprehensive evaluation and explanation of U-Net+ findings as well as its application to X-ray lung segmentation. This research also compares U-Net+ against three additional benchmark segmentation designs and segmentation in identifying tuberculosis and other lung illnesses in the lungs. Our investigation indicates that no previous work has attempted to use U-Net+ for lung segmentation. Due to a lack of segmentation prior to classification, data leakage occurred in the majority of the reviewed articles. Almost no one employed segmentations prior to classification; those who did all relied on U-Net, which U-Net+ can readily supplant because to U-Net+'s superior accuracy with mean iou (described in the findings), which lessen the likelihood of data leaking. Using U-Net+ and a mean iou of 0.95, the authors were able to obtain a lung segmentation accuracy of above 98%, proving the usefulness of this kind of comparison work.