A semi-supervised prototypical network for prostate lesion segmentation from multimodality MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Prostate lesion segmentation from multiparametric MRI is hindered by limited annotated data.
• Supervised models struggle to learn complex features for accurate lesion detection and segmentation due to data scarcity.

Purpose of the Study:

• To develop a novel semi-supervised algorithm for improved prostate lesion segmentation.
• To enhance feature representation for unlabeled data using prototype learning and mean-teacher training.

Main Methods:

• Embedded prototype learning into mean-teacher (MT) training for semi-supervised segmentation.
• Utilized pseudo-labels from the teacher network for unlabeled prototype-based segmentation.
• Enabled bidirectional prototype flow between support and query paths across labeled and unlabeled data.

Main Results:

• The proposed algorithm outperformed state-of-the-art semi-supervised methods on multi-institutional datasets.
• Achieved improved Dice similarity coefficient (0.04-0.09) with increasing labeled data.
• Demonstrated robust performance on the PROSTATEx/PROSTATEx2 datasets as a holdout institute.

Conclusions:

• The novel semi-supervised approach shows significant promise for improving prostate lesion segmentation with limited labeled data.
• This method has the potential to aid clinicians in patient treatment and management decisions.
• The algorithm implementation is publicly available on GitHub.