AI-assisted 3D Anatomical Mapping for Surgical Planning and Education

Thesis Abstract

The advancement of surgical planning and education increasingly depends on accurate, interactive representations of patient-specific anatomy, yet current imaging and visualization approaches often fail to integrate multi-modal data, hinder rapid decision-making, and limit hands-on training. This study investigates AI-assisted 3D anatomical mapping as an integrated platform to enhance preoperative planning, intraoperative guidance, and educational outcomes for health professionals. The aim is to develop and validate a modular AI-driven pipeline that fuses multimodal imaging, quantitative anatomical metrics, and interactive 3D visualization to support personalized surgical strategies and competency-based education. Specific objectives are (1) to design a data fusion framework that integrates CT, MRI, and ultrasound into coherent 3D anatomical maps with automated segmentation; (2) to evaluate the accuracy and reliability of AI-generated models against expert-ground-truth segmentations using Dice similarity, Hausdorff distance, and surface-to-surface error metrics; (3) to assess the impact of the 3D maps on surgical planning times and decision quality in simulated and real-world cases; (4) to examine educational benefits for residents and fellows through structured tutorials, objective structured clinical examinations (OSCEs), and repeated-measures assessments; and (5) to explore ethical, legal, and workflow implications of AI-assisted mapping in clinical practice. The research adopts a mixed-methods design within a pragmatic framework, combining a quantitative validation study with qualitative and mixed-methods evaluations. The population includes four clinical domains—neurosurgery, orthopedics, otolaryngology, and cardiac surgery—encompassing 120 de-identified patient datasets (30 per domain) and 40 trainee participants across three levels of training. Data collection instruments comprise (a) a multi-omics-like imaging repository and labeled ground-truth datasets created via expert consensus and semi-automatic refinement; (b) a validated AI segmentation module based on a 3D U-Net and transformer-augmented post-processing, trained on 80% of the datasets and tested on the remaining 20%; (c) performance metrics including Dice coefficient, Jaccard similarity, and 95th percentile Hausdorff distance; (d) time-motion logs and decision quality checklists captured during standardized surgical planning tasks and simulated operating room scenarios; and (e) educational assessment tools incorporating OSCE scores, knowledge tests, and user satisfaction surveys. Data analysis employs a combination of statistical and machine learning techniques descriptive statistics to characterize datasets, paired t-tests and repeated-measures ANOVA to compare planning efficiency and decision accuracy pre- and post-intervention, Bland-Altman plots to assess agreement between AI maps and expert contours, and multivariate linear mixed-effects models to identify predictors of planning performance. In addition, thematic analysis, grounded in the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT), will be conducted on qualitative interviews with clinicians and educators to elucidate adoption drivers and barriers. The study anticipates that the AI-assisted 3D mapping system will achieve Dice scores above 0.85 and Hausdorff distances below 6 mm for major anatomical structures, reduce planning time by 20–30% in most domains, and significantly improve decision accuracy in complex cases. Educational outcomes are expected to show statistically significant gains in OSCE performance and diagnostic knowledge, with high user satisfaction and perceived usefulness. The anticipated theoretical contribution includes operationalizing the integration of multi-modal data fusion, AI-driven 3D visualization, and outcome-oriented medical education within a single framework, advancing knowledge of enablers and barriers to AI adoption in surgical practice. Practically, the research offers a tested pipeline for scalable deployment in tertiary care centers, with a roadmap for regulatory compliance, data governance, and integration with existing surgical navigation systems. The main conclusion is that AI-assisted 3D anatomical mapping can meaningfully enhance preoperative planning, intraoperative decision support, and competency-based education when embedded in iterative validation, clinician co-design, and rigorous ethical oversight. Recommendations include establishing standardized validation protocols, developing domain-specific educational modules, ensuring transparency in AI predictions through interpretable models, and implementing continuous monitoring to track clinical impact and user experience.