AutoSeg Evaluator

AutoSeg Evaluator reduces the technical and resource barriers required to conduct large-scale segmentation quality evaluations. The tool is freely offered to expedite the assessment and implementation of AI-based autocontouring systems, while also being a useful tool in the pocket of researchers that routinely investigate segmentation quality. AutoSeg Evaluator: • Allows clinicians to overcome the technical expertise barrier required for converting RTSS DICOM files and evaluating segmentations by packaging all functionality within an executable graphical user interface (GUI). • Provides users with a choice of common and advanced metrics using validated open-source implementations. Metrics include volumetric Dice Coefficient, Hausdorff Distance (95th and 100th percentile), Mean Surface Distance, Added Path Length (APL), and Surface Dice. • Supports visualization functionality allowing users to verify outliers and perform sanity checks. • Efficiently handles batched analyses with minimal user interaction by automatically detecting similarly named structures across patients and RTSS files using a template matching feature. Custom string replacement rules enable standardised structure renaming, improving string-matching efficiency. • Computational queues can be set to run in the background across multiple patients and structures while the user prepares the next batch of data. • Outputs detailed results logs for statistical evaluation.