Man vs. Machine (Learning): Comparing Speed and Cost-Effectiveness of Manual vs. Deep Learning Segmentation-Aided Volume Delineation for Head and Neck Lymph Node Targets.

Researchers

Journal

Modalities

Models

Abstract

The purposes of the following study were to: 1) compare the time required to delineate head and neck lymph nodal volumes using a manual versus deep learning segmentation (DLS) – aided approach and 2) investigate the potential cost-savings per scan.10 de-identified head and neck CT simulation datasets were randomly selected for segmentation and analysis. Using published consensus definitions of 27 total lymph node regions1, 1 radiation oncologist manually segmented all 10 CT datasets. For each patient, the time required to contour and review the dataset was recorded. Next, for each dataset, the time required to generate and modify DLS-generated volumes via a proprietary software was recorded, including image export from treatment planning system (TPS), uploading to DLS software, volume generation, downloading volumes, importing volumes into the TPS, and manual edits required. Mean and standard deviations for each treatment modality were calculated. Using two-sided t-test, the statistical difference between the measured outcomes were compared. Finally, assuming a salary of $500,000 per year, average cost savings per scan were calculated.Time required to segment each dataset manually (mean = 59.32 min, SD = 9.16) versus DLS-aided (mean = 14.55 min, SD = 3.16) was statistically significant (p-value = 1.91E-07). Estimated cost required to segment each dataset manually (mean = $237.67, SD = $36.69) versus DLS-aided (mean = $58.27, SD = $12.64) was also statistically significant (p-value = 1.91E-07). The DLS-aided approach resulted in a $179.40 savings per CT dataset segmented, as well as an absolute time savings of 44.78 minutes per scan.This preliminary study shows that using a DLS-aided approach to head and neck lymph node stations can be feasibly implemented within a standard radiation oncology workflow and has the potential for significant time and cost savings.Copyright © 2023. Published by Elsevier Inc.