Development and Comparative Evaluation of Computer-based Assessment Tools for Manual and Robotic Microsurgery

Viola A. Stögner, MD¹, Kai J. Wessel, MD², Catherine T. Yu, BS¹, Sam Boroumand, BS¹, Xiny Xie³, Hannah Ji³, Alex Shi³, Alex Wong, PhD³, Lioba Huelsboemer MD¹, Bohdan Pomahac, MD¹, Tobias Hirsch, MD², Maximilian Kueckelhaus, MD², Haripriya S. Ayyala, MD¹

¹Department of Surgery, Division of Plastic and Reconstructive Surgery, Yale New Haven Hospital, Yale School of Medicine, New Haven, CT, USA

² Department of Plastic and Reconstructive Surgery, Institute of Musculoskeletal Medicine, University Hospital Muenster, Germany

³Yale Vision Lab, Department of Computer Science, Yale University, New Haven, CT, USA

 

Background

Microsurgical evaluations are invaluable

for microsurgical skill development,

however, time- and personnel demanding.

The concept of robotic-assisted

microsurgery raises microsurgical training

requirements of un-/experienced

microsurgeons to a new level. This study

aimed to develop efficient and reliable

tools for microsurgical self-assessment,

applicable to both conventional and

robotic-assisted microsurgery.

 

Material & Methods

Development of two deep convolutional

neural network-based computer

algorithms:

1. Microsurgical instrument tracking

 video analysis

 supervised and semi-supervised

learning used to train models

2. Anastomosis quality assessment

 trained computer algorithm

 analysis of everted anastomosis

images

 Total training images: 90 manual

and 90 robotic anastomoses (1.0

mm Ø silicone vessels (WetLab,

Japan); 10-0 sutures)

 

Results

Microsurgical Instrument Tracking

Deep learning model provides:

 Trajectory of microsurgical instrument

movements (video)

 Efficacy = total trajectory length/time per

anastomosis

 Out of view counts

 Tremor (absolute deviation from smooth

trajectory)

 Average Deviation Intensity:

 Robotic: 6.4 pixel

 Conventional: 24.6 pixel

 

Anastomosis Quality Assessment

Parameters analyzed by computer algorithm:

 Number of stitches

 Disruptions of anastomosis line

 Unequal distancing between stitches

 Oblique stitches causing distortion

 Visible tears in vessel wall

 

Work in progress

 Internal validation (IV):

 First IV: Mean average precision of

~0.80

 Second IV: in progress

 External validation

 Error definition

 WebApp development & pilot study

 

Conclusion

We developed two completely automated

algorithms for microsurgical self-assessment,

promising to markedly increase accessibility

and practicability of feedback-based

microsurgical training.

Both models are trained and evaluated in

conventional and robotic microsurgical

settings, therefore holding great potential to

further leverage and accelerate robotic

evaluation and training in microsurgery.

 

 

References

Chan W, Niranjan N, Ramakrishnan V. Structured assessment of

microsurgery skills in the clinical setting. J Plast Reconstr Aesthetic Surg.

2010;63(8):1329-1334. doi:10.1016/j.bjps.2009.06.024

Ghanem A, Al Omran Y, Shatta B, Kim E, Myers S. Anastomosis Lapse

Index (ALI): A Validated End Product Assessment Tool for Simulation

Microsurgery Training. J Reconstr Microsurg. 2015;32(03):233-241.

doi:10.1055/s-0035-1568157

Milling R, Carolan D, Pafitanis G, Quinlan C, Potter S. Microtools: A

systematic review of validated assessment tools in microsurgery. J Plast

Reconstr Aesthet Surg. 2022 Nov;75(11):4013-4022. doi:

10.1016/j.bjps.2022.06.092. Epub 2022 Jun 29.

 

Contact: Viola A. Stögner, MD, Associate Research Scientist, viola.stogner@yale.edu