Introduction: Microvascular anastomosis is technically demanding for neurosurgeons. Available assessment scales for determination of microanastomosis proficiency rely on subjective Likert-type surveys that must be completed by an expert. We developed a sensorless hand motion detector based on machine learning technology to quantify surgical technique performance metrics in microanastomosis simulation.
Methods: A hand motion detector was developed using a machine learning core (pretrained on large data set of hand images) capable of tracking 21 palm and fingers landmarks without any type of sensors attached. Operators (different expertise levels) performed surgical anastomosis procedures using synthetic vessels while surgical hand motions were recorded using an external camera in front of the surgical field. Coordinates were determined with time series analyses to quantify economy, amplitude, flow and predictability of motion and produce 3D graphs.
Results: 3.426.859 landmark detections were collected from 5 videos analyzed for 600 seconds each. Detection performance was 27.1 measurements per landmark/second. During 600 seconds, 4 non-experts performed 26 bites ([needle drives+suture pull-through] average 6.5 bites/operator) with mean horizontal excess of motion 7.9 (STD 2.6) seconds/bite, while the expert performed 18 bites with mean horizontal excess motion 1.2 (STD 1.1) seconds/bite. In 180 seconds, the expert performed 7 bites, with latency 22.2 (STD 4.4) seconds, while 2 intermediates performed 9 bites (average 4.5 bites/intermediate) combined with mean latency 31.1 (STD 14.6) seconds. Expert time series showed improved predictability over the intermediate group: mean absolute percentage error 5.1% for expert and 10.8% (STD 4.4) for intermediates, the root-mean squared error 30 for expert, 67.6 (STD 39.8) for intermediates.
Conclusion : Surgical technique hand motion detection based on machine learning technology allows features of technical dexterity to be identified, described, and quantified using these performance metrics. Technical expertise could be inferred using this system with implications for surgical skills training and curriculum development.
