The Medical Segmentation Decathlon without a Doctorate.

Segmentation of anatomical structures leverages medical imaging technology for accurate diagnosis and treatment planning. Recent advances in active learning offer the opportunity for the interactive creation of artificial intelligence (AI) segmentation algorithms. Although active learning has been extensively investigated retrospectively, prospective studies have not been widely reported. This study aimed to determine whether a novice in medical imaging could independently label anatomical structures using Radiopaedia, 20% of clinician-generated labels for education of the novice, and AI guidance. Given the pancreas and spleen challenges from the Medical Segmentation Decathlon (MSD), a novice employed the MONAI Label extension on 3D Slicer with active learning (AL) techniques to perform segmentation. This study also explored the use of MONAI (Medical Open Network for AI) Label, a user-friendly framework for label submission and model training. Our segmentation results yielded Dice scores comparable to the range of those from the 2018 segmentation challenges of the MSD. We achieved Dice scores of 0.209 for the pancreas task using 70 images and 0.831 for the spleen using 34 images. The pancreas results do not fall within MSD's reported ranges for the L1 region which is 0.48-0.80. However, the spleen results fall within the MSD's reported range of 0.83-0.97. The interactive AI framework by MONAI Label effectively segments as we reached reasonable performance in less than a month with one person without clinical expertise.