Combining Deep Learning And Medical Knowledge to Detect Cadiomegaly and Pleural Effusion in Chest X-rays Diagnosis

X-ray imaging plays a crucial role in diagnosing various medical conditions, especially those affecting the respiratory and cardiovascular systems. However, interpreting X-ray images can be time-intensive for radiologists. This paper addresses this challenge by developing algorithms to assist radiologists in identifying two specific anomalies in chest X-ray images: cardiomegaly and pleural effusion. Additionally, a key focus is to enhance the understandability and trustworthiness of AI-generated results for medical professionals. To achieve this, we merge deep learning techniques with medical expertise to transform the detection of cardiomegaly and pleural effusion in chest X-rays. We introduce precise U-Net-based segmentation algorithms that delineate critical structures like the heart, lungs, and diaphragm. Furthermore, we propose a novel algorithm to calculate the cardiothoracic ratio, improving cardiomegaly detection accuracy. We also present a method for measuring costophrenic angles to aid in pleural effusion diagnosis and introduce the innovative pneumophrenic contact rate concept for assessing pleural effusion severity. Our performance evaluations reveal superior results compared to the YOLOv5 model, with precision/recall rates of 78%/93% for cardiomegaly and 72%/93% for pleural effusion. This research advances chest X-ray diagnostics, promising more precise disease identification and facilitating AI integration into clinical practice.