2024 World Conference on Lung Cancer (WCLC) - Posters-P4.04C.03 Nodule Type Classification for Lung Cancer Screening with CT

Pdf Summary

The study focuses on enhancing lung nodule classification using AI for lung cancer screening, addressing the challenges with traditional methods that rely on subjective assessments by radiologists. The American College of Radiology's Lung-RADS updates highlight the importance of correctly classifying nodule types for effective patient management. Traditional methods, which compare nodules to adjacent pulmonary vessels, can lead to variability in interpretations, thus demonstrating the need for more objective, automated solutions.<br /><br />The research employs VUNO Med-LungCT AI, an AI-based CADx system, leveraging handcrafted features combined with machine learning to classify lung nodule types accurately. This AI system offers a potential tool for early lung cancer screening by employing Explainable AI (XAI) techniques to clarify which features are critical for classification. The study's findings suggest that model outcomes can be intuitively understood, showing tangible benefits over conventional deep learning models.<br /><br />Machine learning models were trained and tested using a dataset of 1885 nodules, classified by two radiologists. Excluding part-solid and segmentation-failed nodules, the study maintained a focus on other types. The Random Forest algorithm was used to develop the classification model, achieving high sensitivity for both ground-glass (90%) and solid nodules (95.3%).<br /><br />Future research aims to develop models for classifying additional nodule types, such as part-solid and cystic nodules, to enhance the clinical utility further. The study demonstrates the effectiveness of AI in lung nodule classification, providing insights into critical feature importance in the model's framework. This progression towards more accurate and automated classification methods can significantly influence early detection and treatment planning in lung cancer care.

Asset Subtitle

Doohyun Park

Meta Tag

Speaker Doohyun Park

Topic Screening & Early Detection

Keywords

lung nodule classification

AI

lung cancer screening

Lung-RADS

VUNO Med-LungCT AI

Explainable AI

machine learning

Random Forest

ground-glass nodules

solid nodules