2023 World Conference on Lung Cancer (Posters)-EP04.01. AI-based Detection on Low-Dose CT: A Focus on Augmenting Model Performance

Pdf Summary

The study focuses on the characteristics of pulmonary nodules that were misclassified by an artificial intelligence (AI) based computer-aided detection (CAD) system on low-dose chest CT scans. Lung cancer is a leading cause of cancer-related deaths worldwide, and CT scans have proven effective in screening and diagnosing the disease. AI has been extensively researched and applied to CAD systems to assist radiologists in detecting pulmonary nodules and reducing their workload.<br /><br />The study aimed to evaluate the CAD system's performance in detecting pulmonary nodules and analyze the characteristics of false positive and false negative nodules. A dataset of pulmonary nodules was established, and CAD software called VUNO Med -LungCT AI, trained on the LUNA16 dataset, was used for comparison with the readers' results. Statistical measures such as accuracy, sensitivity, and specificity were obtained for both nodule and scan-level assessments.<br /><br />The CAD system achieved sensitivity of 0.72, specificity of 0.70, and accuracy of 0.71 per nodule, and sensitivity of 0.76, specificity of 0.81, and accuracy of 0.78 per case. The study found that the CAD system significantly improved detection performance when used alongside human readers.<br /><br />The study also analyzed the imaging characteristics of false negative and false positive nodules. False negative nodules were typically located peripherally and close to the mediastinum, had ground-glass opacity, fuzzy margins, and dimensions less than 5mm. False positive nodules were often located in the apex and at the convergence of vasculature, and had patterns such as subpleural atelectasis and thickened pleura.<br /><br />In conclusion, the study demonstrated the promising role of deep learning-based CAD systems in nodule detection and the potential for improvement through the application of super-resolution algorithms and tailored training. Further research and development in AI-based CAD systems can enhance the accuracy and efficiency of lung nodule detection, leading to improved lung cancer diagnosis and patient outcomes.

Asset Subtitle

Soo-Youn Ham

Meta Tag

Speaker Soo-Youn Ham

Topic Screening & Early Detection: Biomarkers/Imaging Technology

Keywords

pulmonary nodules

artificial intelligence

computer-aided detection

low-dose chest CT scans

lung cancer

CAD system

false positive nodules

false negative nodules

deep learning-based CAD systems

lung nodule detection