Feasibility evaluation of novel AI-based deep-learning contouring algorithm for radiotherapy

Area of Science:

• Radiotherapy
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate delineation of organs at risk (OARs) is crucial for effective radiotherapy planning.
• Manual contouring of OARs is time-consuming and subject to inter-observer variability.
• AI-driven auto-contouring tools aim to improve efficiency and consistency in radiotherapy workflows.

Purpose of the Study:

• To evaluate the clinical feasibility of the Siemens Healthineers AI-Rad Companion Organs RT (Organs-RT) auto-contouring algorithm.
• To assess the accuracy of Organs-RT auto-contours for OARs in the pelvis, thorax, and head and neck (H&N) regions.
• To measure the time-saving efficiency of the Organs-RT algorithm compared to manual contouring.

Main Methods:

• Computed tomography (CT) datasets from 30 patients (10 pelvis, 10 thorax, 10 H&N) were used.
• Organs-RT generated auto-contours, compared against three independent expert human users.
• Contour accuracy was evaluated using Dice Similarity Coefficient (DSC), Hausdorff Distance (HDD), Mean Distance to Agreement (mDTA), and visual inspection by a physician.
• Time-saving efficiency was measured by recording manual delineation times.

Main Results:

• High agreement (DSC ≥ 0.92) was observed for OARs like the bladder, heart, lungs, and femoral heads.
• Poorer agreement (DSC ≤ 0.81) was noted for the rectum, esophagus, and lips.
• Time-saving efficiency was 67% for H&N, 83% for pelvis, and 84% for thorax.
• Clinical usability scores indicated that 72.5% (pelvis), 82% (thorax), and 50% (H&N) of Organs-RT contours were deemed usable with minimal or no edits.

Conclusions:

• The Organs-RT algorithm shows clinical feasibility as an auto-contouring tool in radiotherapy.
• It effectively minimizes contouring time and enhances time-saving efficiency.
• While agreement varies by OAR, the tool offers significant benefits for radiotherapy treatment planning.