Research on Visual Evaluation of Engineering Skills in Health Applications Based on Artificial Intelligence Assisted 3D Modeling.

Abstract

The artificial intelligence (AI)-assisted 3-D modelling has become central to modern healthcare, yet the field still lacks a repeatable and scalable way to evaluate engineering competence for clinically used digital models. Existing quality checks tend to focus on final mesh outputs without considering the modelling workflow, operator's behavior, and interactions with AI assistance. This gap holds back reproducibility, effective training, and limits regulatory compliance. This work proposes the first end-to-end visual analytics framework that is designed to measure and communicate engineering skill during health-centric 3-D modeling tasks. The proposed framework defines a domain-specific construct of skill, ranging from geometric accuracy to operational proficiency and cognitive adaptability, and quantifies these dimensions through a set of interpretable behavioral, geometric, and physiological indicators. It integrates, within a four-layer architecture, the capture of multimodal data, real-time feature extraction through lightweight deep-learning models, Bayesian evidence fusion for continuous competence estimation, and intuitive visual feedback modules. The system maintains calibration for new tools and modelling scenarios through an online active-learning mechanism that minimizes expert annotation requirements. The framework was tested with 60 participants performing two clinically realistic modeling tasks concerning the design of orthopedic implants and vascular reconstruction. Results showed strong agreement with expert assessments, clear discrimination between skill levels, and meaningful prediction of future modeling quality. Python was used for correlation, regression, validity testing, and visualization tasks, respectively. Usability testing indicated high acceptance; the participants valued highly the clarity of the visual feedback that supported self-directed improvement. Ultimately, this proposed framework reshapes 3-D modelling assessment from a static outcome in-section to a dynamic, process-centric review underpinned by multimodal evidence. The result is a pragmatic, interpretable, and scalable solution for training, certification, and regulatory oversight of health-critical 3-D modelling workflows, and the basis for future human-AI collaboration and competency analytics in medical design.