Optimal selection of candidate datasets for deep learning model retraining using radiomics: application to pancreas segmentation in MRI

Motivation: Machine learning (ML) models need to be periodically evaluated to combat ‘data drift’, where the target population changes over timeGoal(s): The goal is to present a framework for optimally selecting new datasets and updating ML modelsApproach: We retrained a pancreas segmentation model in MRI scans. We selected 50 new cases to annotate using radiomics features, i.e., those unlabelled cases with most differing segmentations from those in the previous training setResults: The system identified a diversity of failure cases, which flagged challenges in real-world data. The mean performance of the model improved after retraining with the additional casesImpact: The proposed system yields a helpful guide for researchers and technicians for retraining machine learning models, particularly deep learning models for organ segmentation in MRI. Selecting an optimal new set of data to annotate produces time and cost savings