AI-Driven 3D Anatomical Mapping for Personalized Surgical Planning

Thesis Abstract

The rapid advancement of artificial intelligence and high-fidelity imaging has driven a paradigm shift in surgical planning, enabling individualized, geometry-aware strategies that adapt to patient-specific anatomy. Despite substantial improvements in image acquisition and segmentation, clinical adoption remains constrained by limited integration of multi-modal data, variability in anatomical representation, and challenges in translating 3D mappings into actionable surgical guidance. This study aims to develop, validate, and evaluate an AI-driven 3D anatomical mapping framework to support personalized surgical planning, with emphasis on improving accuracy, reliability, and decision-support efficacy across diverse patient populations. The specific objectives are (i) to design a multimodal 3D mapping pipeline that fuses CT, MRI, and intraoperative imaging to generate patient-specific anatomical atlases with quantified uncertainty; (ii) to implement deep learning-based segmentation and surface reconstruction algorithms that preserve topological fidelity for critical neurovascular and musculoskeletal structures; (iii) to integrate biomechanical modeling and functional annotations into the 3D maps to assess surgical corridors, residual risk, and tissue handling constraints; (iv) to develop an interactive planning interface that translates 3D maps into quantitative metrics, such as optimal entry trajectories, resection volumes, and predicted intraoperative tissue deformation; and (v) to evaluate the framework’s impact on planning accuracy, decision confidence, and potential operative outcomes using a mixed-methods approach. A mixed-methods research design is applied. The quantitative strand employs a retrospective cohort of 240 anonymized patient imaging datasets drawn from three tertiary centers, with diverse indications including cranial, spinal, and orthopedic procedures. Ground-truth segmentations provided by expert clinicians serve as reference standards. The AI pipeline integrates convolutional neural networks for multi-structure segmentation, geometric deep learning for surface mesh generation, and probabilistic voxel-occupancy models to quantify uncertainty. Biomechanical simulations use finite element analysis to estimate tissue deformation under planned interventions. The qualitative strand includes semi-structured interviews with 18 surgeons to assess perceived usefulness, trust in AI-generated maps, and integration into existing workflows. Data collection instruments include standardized imaging protocols, a validation rubric for segmentation accuracy (Dice similarity coefficient, Hausdorff distance), and a usability questionnaire anchored in the System Usability Scale. Data analysis employs multiple methods. Descriptive statistics summarize map accuracy and uncertainty metrics. Inferential analyses compare planning accuracy against conventional methods using paired t-tests and repeated-measures ANOVA, with effect sizes reported. Regression analyses examine the relationship between map quality (segmentation accuracy and uncertainty) and planning outcomes such as planned resection margins and predicted intraoperative time. Thematic analysis of interview transcripts identifies barriers and facilitators to adoption, guided by seven pre-specified codes derived from the Technology Acceptance Model and Diffusion of Innovation theory. A conceptual model is tested to evaluate the pathways by which AI-driven mapping influences decision quality and surgical risk assessment. Expected findings indicate that the integrated 3D mapping framework reduces planning errors by 22–28% in critical anatomy delineation, improves accuracy of proposed surgical corridors by 15–20%, and lowers predicted intraoperative tissue injury risk markers by 10–15% relative to standard planning. Uncertainty quantification is anticipated to correlate with higher surgeon confidence and greater acceptance of AI-assisted plans. Qualitative insights are expected to reveal that seamless data fusion, transparent uncertainty visualization, and responsive interaction capabilities are pivotal to adoption, while workflow disruption and data governance concerns emerge as key barriers. The study contributes to knowledge by bridging AI-driven imaging analytics with practical surgical planning, providing a validated, uncertainty-aware 3D anatomical mapping system that demonstrates tangible improvements in planning precision and surgeon decision-making. It advances theoretical understanding by empirically testing the integration of probabilistic modeling, biomechanical simulation, and human–machine collaboration in a complex clinical task, framed within the principles of the Technology Acceptance Model and diffusion theory. Practical implications include guidelines for multi-institutional deployment, standards for data interoperability, and recommendations for incorporating AI-driven maps into surgical planning protocols. The main conclusion anticipates that patient-specific 3D anatomical maps, when coupled with robust uncertainty quantification and clinician-centered interfaces, enhance planning quality and may contribute to improved operative safety and outcomes; recommended next steps involve prospective trials, integration with intraoperative navigation systems, and exploration of real-time updates as intraoperative imaging becomes more pervasive.