Assessment of Open Surgery Suturing Skill: Image-based Metrics Using Computer Vision

Abstract

Objective: This paper presents a computer vision algorithm for extraction of image-based metrics for suturing skill assessment and the corresponding results from an experimental study of resident and attending surgeons.

Design: A suturing simulator that adapts the radial suturing task from the Fundamentals of Vascular Surgery (FVS) skills assessment is used to collect data. The simulator includes a camera positioned under the suturing membrane, which records needle and thread movement during the suturing task. A computer vision algorithm processes the video data and extracts objective metrics inspired by expert surgeons' recommended best practice, to "follow the curvature of the needle."

Participants and results: Experimental data from a study involving subjects with various levels of suturing expertise (attending surgeons and surgery residents) are presented. Analysis shows that attendings and residents had statistically different performance on 6 of 9 image-based metrics, including the four new metrics introduced in this paper: Needle Tip Path Length, Needle Swept Area, Needle Tip Area and Needle Sway Length.

Conclusion and significance: These image-based process metrics may be represented graphically in a manner conducive to training. The results demonstrate the potential of image-based metrics for assessment and training of suturing skill in open surgery.

Keywords: image-based metrics; medical simulator; objective metrics; open surgery; skill assessment; suturing.

Figure 1:

Suturing Membrane Housing a) side view of the membrane housing and membrane at surface level, b) top view of the membrane housing, c) membrane at depth level

Figure 2:

The algorithm consists of two stages: In the Image Processing Stage, the needle and thread are detected and needle entry and exit points are identified. In the Metrics Extraction Stage, metrics were computed based on information from the Image Processing Stage.

Figure 3:

Computer vision frame-by-frame display of A) needle and thread detection along with the entry and exit points; B) needle tip path (in red), and swage path (in yellow); C) needle swept area (in black) with the needle detection (in green)

Figure 4:

Needle Tip Trajectory: Pixel list of the needle tip filtered and weeded to be used in post-processing to compute performance metrics.

Figure 5:

A plot of the sway length $l_{sw}(t)$ for one suture cycle, along with example images illustrating the needle with positive (blue) and negative (green) orientation.

Figure 6:

An example of one term in the summation (8) to find Needle Tip Area. The shaded area represents the area added to the summation of index i.

Figure 7:

Frame-by-frame illustration of (A) Needle position (in gray); (B) Cumulative area swept out by the needle body from first frame to current frame (in blue); and (C) Needle tip trajectory from first frame to current frame (in black dashed line). Needle Swept Area metric is the area of the blue in the last frame. Needle Tip Path Length metric is the entire length of the black dashed line in the last frame. Needle Tip Area metric is the total area between the straight-line from entry to exit (red dashed line) and the needle tip path (in green) in the last frame.

Figure 8:

Needle Swept Area and Needle Tip Trajectory Exemplars for an Attending and a Resident

Figure 9:

Experimental results: * indicates statistical significance for $\text{p}<0.05$. (On each box, the middle line indicates the median, and the bottom and top edges of the box indicate the 25^th and 75^th percentiles, respectively. The whiskers are extended to the most extreme data points including outliers.)