Temporal Monte Carlo Dropout for Robust Uncertainty Quantification: Application to Point-of-Care Ultrasound-guided Nerve Blocks

Nishanth Thumbavanam Arun, Leonard Weiss, Andrew Schoenling, Marek Radomski, Frank Guyette, Napoleon Roux, Brittany Daley, Michael J Morris, Howie Choset, John Galeotti

Keywords: POCUS AI, Bayesian Inference, Anatomic Segmentation, PNB, Needle, Nerve Block

TL;DR: We propose a new computationally efficient framework for uncertainty quantification and validate our approach on a point-of-care ultrasound dataset

Abstract: Accurate needle placement during nerve block procedures is essential for safe and effective anesthesia and pain management. However, tracking needles and nerves in an austere setting using Point of Care Ultrasound (POCUS) can be challenging due to the complexity of the surrounding anatomy, the lack of real-time feedback and limited image quality. In this paper, we propose a method for segmenting these structures and estimating the pixelwise uncertainty using a novel approach: Temporal Monte Carlo Dropout. We demonstrate the effectiveness of our approach in POCUS with a stable probe, where it provides robust uncertainty estimates in challenging imaging scenarios while simultaneously tracking the needle accurately. Our method obtains an 84% similarity score with uncertainty estimates obtained from Monte Carlo Dropout with an 8x decrease in computational complexity without compromising segmentation performance. Importantly, it can be easily integrated into existing POCUS workflows on portable devices and has the potential to benefit medical practitioners and patients alike.