The Medical Segmentation Decathlon without a Doctorate

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Accurate segmentation of anatomical structures is crucial for medical diagnosis and treatment planning.
• Active learning (AL) shows promise for developing AI segmentation algorithms interactively.
• Prospective studies on AL for medical image segmentation are limited.

Purpose of the Study:

• To assess if a novice could independently perform anatomical structure segmentation using AL and AI guidance.
• To evaluate the effectiveness of the MONAI Label framework for segmentation tasks.
• To compare segmentation performance against established benchmarks like the Medical Segmentation Decathlon (MSD).

Main Methods:

• A novice utilized the MONAI Label extension on 3D Slicer with AL techniques for segmentation.
• The study focused on pancreas and spleen segmentation challenges from the MSD.
• Radiopaedia and clinician-generated labels were used for novice education and AI model training.

Main Results:

• Segmentation yielded Dice scores of 0.209 for the pancreas (70 images) and 0.831 for the spleen (34 images).
• Spleen segmentation results fell within the MSD's reported range (0.83-0.97).
• Pancreas segmentation results were below the MSD's L1 region range (0.48-0.80).

Conclusions:

• The MONAI Label framework with AL enables effective interactive segmentation.
• Reasonable performance was achieved by a non-expert in under a month.
• This approach highlights the potential for efficient AI-driven segmentation with minimal clinical expertise.