Pulmonary Nodule Detection and False Acceptance Reduction: Review

摘要

Currently, lung cancer alone causes around 20% of all mortalities in cancer. In 2018 alone, 1.8 million deaths were accounted. Lung cancer does not show any prominent symptoms. The survival rate of affected persons can be enhanced by early detection of up to five years. Radiologists examine the chest CT scans slice wise for malignancy detection. The potential challenge in the prediction of the lung nodules is the variation in size, shape, the location of nodules, and the existence of nodules resembling objects. A manual detection is cumbersome and requires a high degree of skill and precision. Researchers throughout the world are currently working on developing a computer-aided approach to assist the radiologists. This study is intended to explore the different techniques for nodule detection, classification, and False Acceptance Reduction (FAR). This paper summarizes the advancements in feature-based and neural network based techniques along with their limitations and required further investigations. This study also highlights and compares a few evaluation metrics used for the assessment of the current methods.