Data Capture in Fluoroscopically-Guided Interventions: A Pathway to Workflow and Procedure Optimization

Purpose: Fluoroscopically-guided interventions (FGIs) offer an interesting link between medical imaging and more traditional procedural interventions such as surgery. The data streams associated with FGIs straddle both arenas, offering potential insights into procedure optimization both within the space of image-guided interventions and other areas of medicine. This work outlines the initial development and application of methods to capture and analyze data streams in FGI procedures. Methods: Data sources included FGI procedure room video feeds, radiation dose structured reports (RDSR), and the electronic medical record (EMR). EMR data was used to identify and categorize procedures of interest via dictation report searches. Procedure videos were identified, matched with EMR data, and then initially processed to remove Private Health Information (PHI). RDSR data was also matched with EMR data and processed/edited for further use. Videos were then converted to still image frames and each frame was classified as an irradiation event (yes/no) and event type if yes. Analyzed video data was matched up with RDSR data according to each irradiation event, and representative images for each event were collected. The level of x-ray field collimation was computed for each event's first and last image frames, and the variations in collimation usage were assessed for both random and directed samplings of procedures. Results: Automated pipelines succeeded in converting procedure videos into sequential image frames, while simple image mask and intensity analysis enabled consistent categorization of each frame by radiation event type. RDSRs only capture information for each irradiation event, which may involve many individual frames (images) and may not capture changes in acquisition parameters during an event. Frame-by-frame analysis of procedure videos identified such changes, adding to the available information from RDSRs alone. Misalignments between events in the RDSR and procedure videos were rare but could be corrected by matching their respective timelines. Analysis of the combined RDSR and video datasets revealed patterns that might be used to identify specific segments of a procedure or specific types of procedures using only the RDSR data. Collimation usage was found to be highly variable across randomly sampled procedures, but much more consistent and aggressive in specific procedures and segments such as embolization sequences in uterine artery embolizations. Conclusion: Methods were developed to capture and analyze key data streams in FGI procedures. The strategy outlined offers analysis and visualization of all FGI events with respect to anatomy exposed to radiation, x-ray acquisition parameters, and procedure segmentation. Hence, there is value in archiving all the images in FGIs. This pilot methodology shows the feasibility of large volume data acquisition and analysis, which should initiate many opportunities to improve FGI practice via large-scale, data-driven approaches.