On using a Particle Image Velocimetry based approach for candidate nodule detection

Detection and segmentation of candidate lung nodules from diagnostic images are vital steps in any image processing-based Computer-Aided Diagnostic (CAD) system for lung cancer. Computed Tomography (CT) is a commonly used modality for lung cancer screening due to the tissue contrast and anatomical resolution. This work aims to investigate the effectiveness of Particle Image Velocimetry, PIV, as a preprocessing tool for processing the input data frames. This is done by applying PIV processing to the input images and quantifying the nodules detected over a morphology-based image processing pipeline. Further, PIV processed images and images without PIV processing were input to the Convolution-based deep learning framework, and the candidate nodule detection effect was quantified and compared. The results validate the efficacy of the proposed workflow for candidate nodule detection both in the image processing pipeline and in the deep learning-based framework. Further, the work also presents the utility of the proposed preprocessing scheme through its ability to detect candidate nodules comprising the major nodule types, namely juxta-pleural, juxta-vascular, isolated, and ground-glass opacity nodules.