Lay summary
Breast surgery aims to remove the tumour while preserving as much healthy breast tissue as possible. A major challenge is that the tumour can shift between diagnostic imaging and the operating room because the breast is soft and changes shape with posture and surgical handling. This uncertainty can contribute to additional surgery, prolonged treatment, distress for patients and whānau, and avoidable cost for the health system. We will develop a marker-less, physics-constrained AI tool that learns how the breast deforms across posture change and during surgery, and provides a 3D estimate of tumour location with an uncertainty range. Unlike data-driven methods, the approach will embed biomechanics so predictions remain realistic and reliable when tissue shape changes. We will test the method using existing imaging and surgical data, and validate performance in controlled breast phantoms to define accuracy, robustness, and safe limits. The outputs will support future clinical evaluation and deployment.