Evaluation of an artificial intelligence U-net algorithm for pulmonary nodule tracking on chest computed tomography images.

OBJECTIVES: To apply image registration in the follow up of lung nodules and verify the feasibility of automatic tracking of lung nodules using an artificial intelligence (AI) method. METHODS: For this retrospective, observational study, patients with pulmonary nodules 5-30 mm in diameter on computed tomography (CT) and who had at least six months follow-up were identified. Two radiologists defined a 'correct' cuboid circumscribing each nodule which was used to judge the success/failure of nodule tracking. An AI algorithm was applied in which a U-net type neural network model was trained to predict the deformation vector field between two examinations. When the estimated position was within a defined cuboid, the AI algorithm was judged a success. RESULTS: In total, 49 lung nodules in 40 patients, with a total of 368 follow-up CT examinations were examined. The success rate for each time evaluation was 94% (345/368) and for 'nodule-by-nodule evaluation' was 78% (38/49). Reasons for a decrease in success rate were related to small nodules and those that decreased in size. CONCLUSION: Automatic tracking of lung nodules is highly feasible.

Development and performance evaluation of a deep learning lung nodule detection system.

BACKGROUND: Lung cancer is the leading cause of cancer-related deaths throughout the world. Chest computed tomography (CT) is now widely used in the screening and diagnosis of lung cancer due to its effectiveness. Radiologists must identify each small nodule shadow from 3D volume images, which is very burdensome and often results in missed nodules. To address these challenges, we developed a computer-aided detection (CAD) system that automatically detects lung nodules in CT images. METHODS: A total of 1997 chest CT scans were collected for algorithm development. The algorithm was designed using deep learning technology. In addition to evaluating detection performance on various public datasets, its robustness to changes in radiation dose was assessed by a phantom study. To investigate the clinical usefulness of the CAD system, a reader study was conducted with 10 doctors, including inexperienced and expert readers. This study investigated whether the use of the CAD as a second reader could prevent nodular lesions in lungs that require follow-up examinations from being overlooked. Analysis was performed using the Jackknife Free-Response Receiver-Operating Characteristic (JAFROC). RESULTS: The CAD system achieved sensitivity of 0.98/0.96 at 3.1/7.25 false positives per case on two public datasets. Sensitivity did not change within the range of practical doses for a study using a phantom. A second reader study showed that the use of this system significantly improved the detection ability of nodules that could be picked up clinically (p = 0.026). CONCLUSIONS: We developed a deep learning-based CAD system that is robust to imaging conditions. Using this system as a second reader increased detection performance.