Evaluating Margin Sharpness Analysis on Similar Pulmonary Nodule Retrieval

Abstract

Lung cancer is the leading cause of cancer-related deaths in the world and its main manifestation is through pulmonary nodules. Pulmonary nodule classification is a challenging task that must be done by qualified specialists, but image interpretation errors and temporal aspects difficult those processes. In order to aid radiologists on the image interpretation process, it is important to integrate computer-based tools with the lung cancer diagnostic process. Content-Based Image Retrieval (CBIR) can provide decision support to specialists by allowing them to find images from a database that are similar to a reference image. However, a well known challenge of CBIR is the image feature extraction process. Margin sharpness descriptors are still imatures and need to be more evaluated in order to optimize the performance of similar pulmonary nodule retrieval. The goal of this work is to perform a Margin Sharpness Analysis (MSA) in pulmonary nodule presented in computed tomography images, to retrieve the most similar nodules based on this MSA and to evaluate the performance of margin sharpness descriptors in the nodule retrieval. The results show that MSA presented a mean precision of 0.62 and 0.63, according to Precision and Recall parameters, regardless nodule malignancy, with Euclidean and Manhattan distances as image similarity measures, respectively. The evaluation also showed that, for the first 10 similar cases, the mean precision was 0.81 for both similarity distances.