Page 73 - Read Online

Page 355 Hogue et al. Art Int Surg. 2025;5:350-60 https://dx.doi.org/10.20517/ais.2025.19

anatomical landmarks on the unilateral cleft lip essential for the placement of nasolabial markings prior to
[21]
cleft lip repair .
In addition, efforts to integrate AI into virtual reality (VR) surgical simulators have been successful.
Software has been developed that can assess surgical skills during a simulation and distinguish novice from
expert users [22,24] . In 2019, Siyar et al. showcased the ability of AI software to distinguish expert versus novice
[22]
performance using a surgical skills simulator . Neurosurgery trainees performed a VR tumor resection
task, and the software assessed surgical trainees using classifiers of operative dexterity such as speed or
applied force. Similarly, Fukuta et al. developed a novel algorithm that can assess forceps manipulation
during surgical simulation . However, the system cannot yet distinguish expert-level use of forceps from
[20]
novice-level use. The authors reported plans to continue improving the AI model’s forceps tracking abilities
and develop a system to train surgeons in laparoscopic skills.

More recent studies show that AI can replace the human tutor. A trial conducted by Lei et al. demonstrated
that the use of an AI-assisted ultrasound system helped residents reach proficiency milestones in fewer
training cycles compared to those receiving standard instruction . Similarly, Yilmaz et al. developed an AI-
[25]
powered software using a long short-term memory (LSTM) network that analyzes sequences of movements
over time to continuously monitor surgical skills and provide real-time feedback as would a human
[24]
instructor . Again, neurosurgical trainees at various levels of training were successfully distinguished by
the algorithm. Within the field of endodontic surgery, Vannaprathip et al. developed the Surgical Decision-
making Mentor (SDMentor) as the first AI-based system to teach surgical decision making, combining a VR
simulator with an AI tutor . The quality of SDMentor feedback was compared against that of human
[23]
tutors, and the AI software performed better. Expert dental instructors were only able to correctly identify
which suggestions were provided by the AI tutor 15% of the time .
[23]

Additionally, a randomized control trial showed that an AI-powered virtual operative assistant can
effectively instruct users during surgical simulation training compared to users receiving expert or no
guidance . Performance was assessed using the AI skill assessment software developed by Yilmaz et al. .
[19]
[24]
The group receiving AI instruction experienced more improved performance than either the expert or
control groups. However, the rate of improvement was similar to that of the group that received expert
instruction. Surveyed participants reported increased positive emotions when receiving AI-generated
feedback similar to that occurring with human feedback.

Improving resident feedback
A cornerstone of resident education is the feedback provided by surgical faculty. In addition to
examination, performance feedback is a primary method by which residents can review and use to improve
their performance. No attempt to improve plastic surgery resident feedback is currently available in the
literature. However, several papers detail novel attempts to use AI to quantify faculty feedback within the
field of general surgery [26-28] .

Stahl et al. used AI to successfully identify key topics among entrustable professional activities (EPA)
[28]
feedback given to general surgery residents . EPAs are micro assessments that standardize competency-
based feedback regarding specific, essential activities of the competent surgeon, designed to facilitate
immediate feedback during the daily workflow. However, performance attributed to each entrustment level
was defined by experts’ opinions, so Stahl et al. studied feedback data to determine what differentiates
entrustment levels . NLP software identified words within the EPA comments and determined which
[28]
corresponded with each entrustment level. Unsurprisingly, surgeons used distinct words and sentence