Machine Learning for Human-Autonomy Teaming in Surgical Skill Assessment: Scoping Review.

BACKGROUND

Human-autonomy teaming (HAT) has the potential to reshape surgical practice by fostering true partnership between surgeons and intelligent systems. To achieve this, AI must move beyond static scoring to provide adaptive, real-time guidance based on reliable skill assessment.

OBJECTIVE

This scoping review aims to map the current landscape of machine learning methods for surgical skill assessment and evaluate their technical readiness for integration into functional HAT systems.

METHODS

Following PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines, we conducted a systematic search across 3 major scientific databases (PubMed, IEEE Xplore, and Web of Science). We identified and analyzed 92 peer-reviewed studies published between 2019 and 2025. The review focused on data modalities (kinematics, video, and biosignals), model architectures, and validation environments. To assess translational maturity, each study was evaluated using the NASA technology readiness level (TRL) scale and mapped to the Yang et al levels of autonomy for surgical robotics, alongside 6 additional dimensions: real-time capability, interpretability, adaptivity, validation environment, external validation, and clinical integration.

RESULTS

Our analysis of the 92 included studies reveals a dominant shift toward multimodal data integration and deep learning architectures. While high performance is frequently reported on benchmark datasets, significant barriers to HAT integration persist. The TRL analysis shows that 77.2% (71/92) of studies remain at TRL 3 (proof-of-concept), with only 2.2% (2/92) reaching TRL 5 or above. Furthermore, 96.7% (89/92) of studies operate at Yang autonomy level 0, providing no autonomous functionality beyond classification. We identified that 91.3% (84/92) of models are static, with no user-specific adaptation, 38% (35/92) lack any external validation, and validation practices remain inconsistent across the field.

CONCLUSIONS

Current AI techniques provide a robust foundation for objective skill assessment, but they are not yet ready for autonomous teaming. Future development must prioritize model robustness, interpretability, and seamless integration into clinical environments to transition from stand-alone assessment tools to effective surgical teammates.

TRIAL REGISTRATION: OSF Registries 10.17605/OSF.IO/PQWS5; https://osf.io/pqws5.