2024 World Conference on Lung Cancer (WCLC) - ePosters-EP.08F.10 Semi-Automated NSCLC Segmentation and RECIST Measurement: Bridging the Gap Between Speed and Radiologist-Level Accuracy

EP.08F.10 Semi-Automated NSCLC Segmentation and RECIST Measurement: Bridging the Gap Between Speed and Radiologist-Level Accuracy

Back to course

Pdf Summary

The study presented by Kai Zhang and colleagues introduces ClickSeg, a semi-automated algorithm designed to improve the efficiency and accuracy of segmentation and measurement of non-small cell lung cancer (NSCLC) lesions. Traditional RECIST (Response Evaluation Criteria In Solid Tumors) measurements, which are crucial for assessing treatment responses, are time-consuming and subject to variability among different observers. ClickSeg addresses these issues by allowing users to click on a lung lesion, which the algorithm then segments automatically and calculates the RECIST sum of target lesion diameters (SLD).<br /><br />The ClickSeg algorithm was trained using data from 1,000 chest CT images and validated with 753 target lesions from 557 patients across four different institutions. The study demonstrated that ClickSeg can perform segmentation and measurement tasks 60 times faster than manual methods, with an average completion time of 14.59 seconds compared to 15.5 minutes manually. The algorithm showed a Pearson’s correlation of 0.81 and an intra-class correlation of 0.85 for SLD measurement, indicating strong agreement with human experts.<br /><br />Notably, despite its efficiency, the algorithm achieved a median Dice coefficient of 0.66, reflecting moderately good segmentation accuracy. The study acknowledges its limitations, including its retrospective nature and lack of inter-rater reliability metrics among human annotators. Future work aims to test ClickSeg in longitudinal studies and further explore its practical integration into clinical settings.<br /><br />Overall, the findings suggest that ClickSeg could significantly enhance the speed and consistency of NSCLC lesion analysis, potentially improving clinical workflows and reducing observer variability in tumor response assessment.

Asset Subtitle

Kai Zhang

Meta Tag

Speaker Kai Zhang

Topic Local-Regional Non-small Cell Lung Cancer

Keywords

ClickSeg

semi-automated algorithm

NSCLC lesions

RECIST measurements

segmentation accuracy

lung cancer

clinical workflows

observer variability

tumor response assessment

Pearson's correlation