Automated Deep Learning-based Detection and Segmentation of Lung Tumors at CT Imaging

Background: Detection and segmentation of lung tumors on CT scans is critical for monitoring cancer progression, evaluating treatment responses, and planning radiation therapy; however, manual delineation is labor-intensive and subject to physician variability. Purpose: To develop and evaluate an ensemble deep learning model for automating identification and segmentation of lung tumors on CT scans. Materials and Methods: A retrospective study was conducted between July 2019 and November 2024 using a large dataset of CT simulation scans and clinical lung tumor segmentations from radiation therapy plans. This dataset was used to train a three-dimensional U-Net multiresolution ensemble model to detect and segment lung tumors on CT scans. Model performance was evaluated on internal and external test sets composed of CT simulation scans and lung tumor segmentations from two affiliated medical centers; scans included a single primary or metastatic lung tumor. Performance metrics included sensitivity, specificity, false-positive rate, and Dice similarity coefficient (DSC). Model-predicted tumor volumes were compared with physician-delineated volumes. Segmentation outcomes were compared using the Wilcoxon signed rank test or one-way analysis of variance, with P < .05 indicating a statistically significant difference. Results: The model, trained on 1504 CT scans with clinical lung tumor segmentations, achieved 92% sensitivity (92 of 100) and 82% specificity (41 of 50) in detecting lung tumors on the combined test set of 150 CT scans. For tumor segmentations on the subset of 100 CT scans with a lung tumor, median model-physician DSC was 0.77 (IQR, 0.65-0.83), and median interphysician DSC was 0.80 (IQR, 0.72-0.86). Segmentation time was shorter for the model than for physicians (mean, 76.6 seconds vs 166.1 and 187.7 seconds; P < .001 for both). Conclusion: Routinely collected radiation therapy data were useful for model training. The key strengths of the model include a three-dimensional U-Net ensemble approach for balancing volumetric context with resolution, robust tumor detection and segmentation performance, and generalizability to an external site.