Validating our computer vision approach against gold standards
We've started the process of benchmarking our Computer Vision patient alignment technique against the current gold standard - IR-marker motion tracking.
In a 20 heathy-volunteer study, we will quantify and benchmark our AI technique using optical only imaging. Each volunteer will participate in a gold standard marker-based motion capture collection in the Interdisciplinary Movement Science Laboratory on the Anschutz Medical Campus.
We record whole-body motion during replication of clinical breast RT positioning during free breathing, thoracic DIBH, and abdominal DIBH. Motion will be captured using 3-D position data from reflective markers measured from 10 infrared cameras (Fs = 100 Hz) (the gold standard for motion tracking).
Simultaneously, we will acquire artificial intelligence (AI) human pose tracking using 6-10 synchronized cameras. The accuracy of our technique will be quantified against the motion capture gold-standard, and the optimum positioning and minimum number of cameras required to match current state of the art positioning accuracy will be identified.
In a second optical-only imaging study, the volunteers will be imaged with current state-of-the-art surface guided imaging, with simultaneous AI human pose tracking will be recorded, and the resulting measurements compared.
This benchmarking study is a critical step in validating our computer vision approach for clinical use. By directly comparing our method against the current gold standard, we can quantify the accuracy, precision, and reliability of our technique. This data will be essential for regulatory approval and clinical implementation, potentially leading to more accessible and effective patient positioning systems for radiation therapy.