Artificial Neural Network Approach to Competency-Based Training Using a Virtual Reality Neurosurgical Simulation

Abstract

BACKGROUND: The methodology of assessment and training of surgical skills is evolving to deal with the emergence of competency-based training. Artificial neural networks (ANNs), a branch of artificial intelligence, can use newly generated metrics not only for assessment performance but also to quantitate individual metric importance and provide new insights into surgical expertise. OBJECTIVE: To outline the educational utility of using an ANN in the assessment and quantitation of surgical expertise. A virtual reality vertebral osteophyte removal during a simulated surgical spine procedure is used as a model to outline this methodology. METHODS: Twenty-one participants performed a simulated anterior cervical diskectomy and fusion on the Sim-Ortho virtual reality simulator. Participants were divided into 3 groups, including 9 postresidents, 5 senior residents, and 7 junior residents. Data were retrieved from the osteophyte removal component of the scenario, which involved using a simulated burr. The data were manipulated to initially generate 83 performance metrics spanning 3 categories (safety, efficiency, and motion) of which only the most relevant metrics were used to train and test the ANN. RESULTS: The ANN model was trained on 6 safety metrics to a testing accuracy of 83.3%. The contributions of these performance metrics to expertise were revealed through connection weight products and outlined 2 identifiable learning patterns of technical skills. CONCLUSION: This study outlines the potential utility of ANNs which allows a deeper understanding of the composites of surgical expertise and may contribute to the paradigm shift toward competency-based surgical training.

Utilizing a multilayer perceptron artificial neural network to assess a virtual reality surgical procedure

“…The ACDF simulated scenario utilized in this study has been extensively employed by our group to assess surgical expertise. The simulation includes 3 animated steps (neck incision, placement of retractors, and fusion) and 4 deconstructed interactive steps (C4-C5 vertebral disc annulus incision, discectomy, osteophyte removal, and posterior longitudinal ligament removal) [5,6,18]. Each of the interactive simulated steps have been shown to have face, content and construct validity [5].…”

“…The current study focuses on the first interactive step which consists of performing a 2cm transverse box incision exposing the disc annulus using a virtual No.15 scalpel. The second interactive step, discectomy, has been assessed by Mirchi, et al [6] and the third interactive step, osteophyte removal by Reich, et al [18] have been previously reported.…”

“…Following the SFS (sequential forward selection) algorithm, only nine important metrics remained, as demonstrated in Table 4. Similar to the data from the discectomy but unlike the osteophyte removal study, the nine most significant metrics spanned all three categories [6,18]. These nine surgical performance metrics were used as inputs to the developed ANN.…”

“…Figure 8 demonstrates a visualization of the CWP trends between the junior, senior, and postresident groups for each feature. Two main learning patterns have been identified using ANN to assess the surgical performance of post-residents, senior and junior residents during the simulated ACDF procedure on the Sim Ortho Platform [6,18]. These two patterns have been identified as continuous and discontinuous learning.…”

“…Each component of the ACDF simulation was previously validated by Ledwos, et al [5]. The second and third steps of the surgical simulation, concerning the discectomy and osteophyte removal components, have been outlined [6,18]. These studies utilized some of the same participant data to generate metrics and extract CWPs from developed ANNs, employing similar methodology.…”