2024 World Conference on Lung Cancer (WCLC) - Posters-P4.07F.06 CT-Based Deep Learning Model for Predicting Visceral Pleural Invasion in NSCLCs with Tumor Diameter Less Than 3cm

P4.07F.06 CT-Based Deep Learning Model for Predicting Visceral Pleural Invasion in NSCLCs with Tumor Diameter Less Than 3cm

Back to course

Pdf Summary

The study explores improving preoperative prediction of visceral pleural invasion (VPI) in early-stage non-small cell lung cancer (NSCLC) by developing a computed tomography (CT)-based deep learning (DL) model. Specifically, it targets NSCLC cases with tumors less than 3 cm in diameter. Accurate VPI prediction is crucial as it significantly influences surgical decisions and outcomes.<br /><br />A total of 835 patients from two institutes were retrospectively analyzed, with 691 patients from institute 1 used for the training and internal validation of the model, and 144 patients from institute 2 for external validation. The novel VPI-DL model achieved a notably high area under the curve (AUC) of 0.91 and an accuracy rate of 83% on the testing set, outperforming traditional approaches and existing methods. In external validation, it showed superior performance compared to the previously established DL model, consolidation-to-tumor (C/T) ratio, and assessments by thoracic surgeons, achieving a higher AUC and better accuracy and specificity.<br /><br />The study underscores the VPI-DL model's capability to offer an expert-level, predictive tool that aids surgical planning considerably better than traditional clinical evaluations. Notably, the model's prediction confidence correlated highly with physicians' interpretations, especially in complex cases, demonstrating that it can effectively assist in decision-making processes.<br /><br />Overall, the new CT-based VPI-DL model represents a significant advancement in the preoperative assessment of NSCLC, promising to improve surgical outcomes by providing more reliable predictions of visceral pleural invasion than current standard practices.

Asset Subtitle

Mong-Wei Lin

Meta Tag

Speaker Mong-Wei Lin

Topic Early-Stage NSCLC

Keywords

visceral pleural invasion

non-small cell lung cancer

CT-based deep learning

preoperative prediction

VPI-DL model

surgical decisions

area under the curve

external validation

predictive tool

surgical outcomes