Software and Automation Algorithms for Chart Rounds: Improving the Peer-to-Peer Review Process

Abstract

Peer-review of treatment plans in chart rounds meetings is a tedious process, as plan quality and compliance to accepted standards is still assessed visually. We have developed algorithms for the automated verification of the plan as well as software with design features aimed at facilitating a comprehensive review of plans in minimal times. The automation is implemented through software modules that are called once a treatment plan is finalized. Checks range from simple automations to verify compliance of prescription and doses against NRG DVH endpoints to a complex statistical shape algorithm modeling inter-patient variability as trained on approved patient plans. This learning algorithm is employed here to verify the correctness of OAR contouring and predict achievable doses for critical structures based on their clearances from the targets. Any deviations from the norm are automatically flagged for review. These automated checks are paired with modern software to visualize the images, contours, and dose in a streamlined fashion during chart rounds meetings. Various practical features have been incorporated in this software, such as recording the meeting’s discussions to document recommendations, touch-screen capabilities, or loading image datasets as a separate software process in the background while the prescription and plan details are discussed by the attendees. The approach has been used clinically since November 2016 for quality assurance of 2029 patient plans across a wide variety of physician preferences and treatment techniques that were easily verified by the proposed method. Review times are kept under 2 minutes per patient, this including a review of the prescription, browsing images to verify contours accuracy, inspection of the DVHs and associated constraints. The most common flagged issue in the prescription, at 150 occurrences out of 176 clerical checks performed by the software, was an unspecified disease staging. Out of 7541 constraints automatically checked, 6233 were easily fulfilled, 279 were borderline, and 1023 were not fulfilled mainly because of OAR-PTV clearances that were documented and visualized with the software. The modules to predict achievable doses and verify OARs geometry though comparison with shape models build from approved plans were valuable tools to justify clinical decisions for these challenging situations. Our department developed automation software for chart rounds meetings that flags inaccuracies in the plan. The core of the system is the comparison of a patient's datasets with a statistical model of accepted geometries learned approved plans that creates means to validate segmentations and dosimetric endpoints. This quality assurance approach streamlines the review process and provides an e-learning experience for the attending’s and residents to discuss cases in a modern software interface designed for this task.