Enhancing Lung Cancer Detection in X-Rays: Advanced Segmentation of Varied Nodule Sizes

Thoracic Computerized Tomography (CT) scans are renowned for their detailed lung structure visualization, crucial for detecting early-stage lung nodules. However, they come with limitations such as high cost, limited accessibility, radiation exposure, and reduced portability. Alternatively, Chest X-rays (CXRs), while more accessible and less costly, struggle in nodule detection due to their two-dimensional nature. This study bridges this gap by leveraging CT scan data to enhance lung cancer detection in CXRs, thereby improving radiologists’ ability to segment lung cancer with greater confidence using CXR images alone. Our approach involves fine-tuning a model on Digitally Reconstructed Radiographs (DRRs) to identify small nodules in CXRs, a notable challenge in current practices. The system demonstrates superior performance over existing methods in terms of mIOU, recall, and precision metrics. Specifically, it achieves an MIOU of 0.9025, recall of 0.9505, and precision of 0.9212, significantly enhancing lung cancer detection capabilities in CXRs. This research represents a significant step in addressing the limitations of CXRs for lung cancer detection, offering a cost-effective, accessible, and efficient alternative to CT scans, particularly in resource-constrained areas. By integrating CT volume features into CXR analysis, our model shows promise in detecting lung nodules of various sizes, potentially transforming the diagnostic process in radiology.