Deep-learning based Tools for Automated Protocol Definition of Advanced\n Diagnostic Imaging Exams

Abstract

Purpose: This study evaluates the effectiveness and impact of automated\norder-based protocol assignment for magnetic resonance imaging (MRI) exams\nusing natural language processing (NLP) and deep learning (DL).\n Methods: NLP tools were applied to retrospectively process orders from over\n116,000 MRI exams with 200 unique sub-specialized protocols ("Local" protocol\nclass). Separate DL models were trained on 70\\% of the processed data for\n"Local" protocols as well as 93 American College of Radiology ("ACR") protocols\nand 48 "General" protocols. The DL Models were assessed in an "auto-protocoling\n(AP)" inference mode which returns the top recommendation and in a "clinical\ndecision support (CDS)" inference mode which returns up to 10 protocols for\nradiologist review. The accuracy of each protocol recommendation was computed\nand analyzed based on the difference between the normalized output score of the\ncorresponding neural net for the top two recommendations.\n Results: The top predicted protocol in AP mode was correct for 82.8%, 73.8%,\nand 69.3% of the test cases for "General", "ACR", and "Local" protocol classes,\nrespectively. Higher levels of accuracy over 96% were obtained for all protocol\nclasses in CDS mode. However, at current validation performance levels, the\nproposed models offer modest, positive, financial impact on large-scale imaging\nnetworks.\n Conclusions: DL-based protocol automation is feasible and can be tuned to\nroute substantial fractions of exams for auto-protocoling, with higher accuracy\nwith more general protocols. Economic analyses of the tested algorithms\nindicate that improved algorithm performance is required to yield a practical\nexam auto-protocoling tool for sub-specialized imaging exams.\n

Ähnliche Arbeiten