Accuracy and reproducibility of tumor size measurement using a deep-learning-based CDSS in resected lung cancer

MONCAD LCT is a commercially available deep-learning based clinical decision support system (CDSS) for lung screening CT. The aim of this multicenter retrospective study was to evaluate the accuracy and reproducibility of tumor size measurement using a commercially available deep-learning-based clinical decision support system (CDSS), compared with radiologist assessments and pathological reference in resected lung cancer cases.

We retrospectively collected preoperative CT images and original radiology reports and the CDSS results for resected lung cancer from three institutions during 2022 (n = 176). MONCAD LCT evaluated the LungRADs category based on the density and size of the lung nodule. First of all, we compared the MONCAD LCT and original radiologic report using the pathologic tumor size as gold standard. Furthermore, the subsampling case (n = 33) randomly selected by institutions, density type (pure ground glass opacity, subsolid, and solid) and tumor size, two expert thoracic radiologists independently evaluated the tumor size for the resected lung cancer and the interobserver variability was evaluated.

Among 176 tumors, 162 (92%) were detected on MONCAD LCT. Tumor size measurement by original radiology report and CDSS were found to have excellent reliability with pathologic tumor size (ICC = 0.869 for absolute agreement). On reader study, excellent interobserver agreement (ICC = 0.907) was observed between two expert radiologists, which was inferior to the completely consistent CDSS results (ICC = 1.000).

No significant differences were found in the measurement of tumor size between radiologists and the CDSS. CDSS might be helpful to minimize interobserver variability for tumor size measurement by supplying consistent and reliable results.

This real-world multicenter study demonstrates that the CDSS provides consistent and objective tumor size measurements, supporting its potential utility in standardizing preoperative lung cancer assessment.