Systematic Clinical Evaluation of a Deep Learning Method for Medical Image Segmentation: Radiosurgery Application

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Manual segmentation in 3D medical imaging suffers from high inter-rater variability and is time-consuming.
• Existing deep learning evaluations often lack detailed analysis applicable to diverse segmentation tasks.

Purpose of the Study:

• To systematically evaluate a Deep Learning (DL) model for 3D medical image segmentation.
• To quantify the model's impact on inter-rater agreement and delineation efficiency.
• To provide insights for developing efficient DL models in this domain.

Main Methods:

• A Deep Learning model was applied to a 3D medical image segmentation task.
• Inter-rater detection and contouring agreements were analyzed before and after model application.
• Delineation time was measured to assess efficiency.
• Clinical experiment design incorporated bias mitigation strategies.

Main Results:

• The DL model reduced detection disagreements by [Formula: see text].
• Inter-rater contouring agreement improved from [Formula: see text] to [Formula: see text] surface Dice Score.
• The model accelerated the delineation process by [Formula: see text] to [Formula: see text] times.
• Bias-controlled clinical evaluation preserved result significance.

Conclusions:

• Deep Learning models offer a viable solution to manual segmentation challenges in 3D medical imaging.
• The evaluated DL model significantly improves accuracy, consistency, and efficiency.
• The study provides practical insights for the development and application of DL in medical image segmentation.