A Proof-of-Concept Study of Artificial Intelligence-assisted Contour Editing.

Area of Science:

• Medical imaging and artificial intelligence
• Radiotherapy and oncology
• Computational pathology

Background:

• Accurate segmentation of organs and tumors in medical imaging is crucial for effective treatment planning.
• Manual contouring is time-consuming and subject to inter-observer variability.
• Deep learning models show promise in automating segmentation tasks but often require refinement.

Purpose of the Study:

• To introduce and demonstrate the feasibility of artificial intelligence-assisted contour editing (AIACE).
• To present a novel workflow where clinicians guide deep learning models for contour refinement.
• To evaluate the efficiency and effectiveness of AIACE in head-and-neck cancer CT imaging.

Main Methods:

• A retrospective proof-of-concept study using 2D axial CT images from three head-and-neck cancer datasets.
• Simulated clinical environment where clinicians provided input via mouse clicks to guide the AIACE model.
• Iterative refinement of contours until clinical acceptability was achieved.
• Quantification of model performance using Dice Similarity Coefficient (DSC) and Hausdorff Distance 95th percentile (HD95).

Main Results:

• Initial automated contours showed average DSCs ranging from 0.67 to 0.82 and HD95 from 4.3 to 11.4 mm.
• AIACE improved contours to average DSCs of 0.86-0.91 and HD95 of 2.1-3.3 mm with approximately three mouse clicks.
• Each deep learning-based contour update was completed in about 20 milliseconds.

Conclusions:

• The proposed AIACE concept effectively assists clinicians in contour editing.
• AIACE demonstrates significant improvements in segmentation accuracy and efficiency.
• The study confirms the feasibility of using deep learning for interactive medical image segmentation in clinical workflows.