Evaluating the Effectiveness of Deep Learning Contouring across Multiple Radiotherapy Centres

Area of Science:

• Radiotherapy and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Anatomy

Background:

• Deep learning contouring (DLC) offers potential for reducing manual organ delineation time and inter-observer variability in radiotherapy.
• Evaluating the clinical utility of DLC across multiple institutions is crucial for its adoption.

Purpose of the Study:

• To assess the effectiveness of a commercial deep learning contouring system for prostate and head and neck radiotherapy.
• To quantify time savings and contour accuracy of DLC compared to traditional methods.

Main Methods:

• Analysis of CT scans from 123 prostate and 310 head and neck cancer patients across four radiotherapy centers.
• Comparison of manual contouring times with DLC editing times using paired and non-paired study designs.
• Evaluation of contour agreement using Dice Similarity Coefficient (DSC) and Distance to Agreement (DTA).

Main Results:

• Mean time savings of 5.9 ± 3.5 min for prostate and 16.2 ± 8.6 min for head and neck structures were observed.
• High agreement (DSC: 0.92 ± 0.03) for femoral heads, with lower agreement for the rectum (DSC: 0.68 ± 0.04).
• DLC contours demonstrated a reduction in inter-observer variability for structures like the brainstem and salivary glands.

Conclusions:

• Generic DLC models for prostate and head and neck are effective in generating time savings, verifiable through various study designs.
• DLC has the potential to reduce inter-observer variability, contributing to more consistent radiotherapy planning.