Computer-aided .detection for pulmonary nodule identification: improving the radiologist's performance?

摘要

Computer-aided detection (CAD) systems for pulmonary nodule identification in CT images can represent valuable tools in assisting radiologists with a second opinion in the detection of early-stage lung cancers. This task becomes more and more challenging when low-dose CT protocols are implemented, which are the optimal choice for screening the asymptomatic population. The absolute performance of the CAD systems, as well as the reliability and reproducibility of the results across different data samples, is a fundamental requirement for these tools to be integrated in the diagnostic algorithm for lung cancer. An overview of the methods recently implemented to build CAD systems and a comparison of the performance achieved are provided. A fair comparison can be carried out only on common data samples, thus the validation of CAD systems on publicly available data sets, such as the Lung Image Database Consortium database, is highly recommended. The necessary steps to evaluate whether these systems can be valuable second readers of CT images are also discussed. The full exploitation of the CAD potential on the accuracy of diagnostic image interpretation requires the integration of the algorithm in the workstations used for image reviewing, and access to a picture archiving and communication system environment. CAD should be accessible, fast and easy to use and maintain.