Markerless Tracking of the Breast

When treating breast cancer, the large soft tissue deformations of the breast can complicate accurate localisation of tumours. This effects initial screening, surgery, and follow up care for patients with breast cancer. While soft tissue mechanics can predict internal deformation, lack of information about the shape and position of the patient and their breast can limit their application. While information about the location of the tumour is relevant at many stages of the treatment of breast cancer, it is particularly relevant in breast conserving surgery. Replacing mastectomy, breast conserving surgery aims to remove just the tumour. A successful breast conserving surgery removes a tumour with a small (approx 2 mm) margin, but uncertainty in tumour shape and position can lead to the need for re-excision or mastectomy. This thesis addresses these challenges by developing a model-based approach for precise tracking of breast surface deformation from stereo imaging. This will allow for validating the predictions of physics-based models in positions that account for the changing position of the arm and shoulder. Ultimately, these measurements and models will allow us to convey information for interventions: where a tumour is; how its shape has changed; and how it is embedded within the soft tissue landscape of the breast. This information will help clinicians ensure they accurately find the boundaries of tumours, reducing re-operation rates for breast conserving surgery. The thesis describes three methodological contributions that provide the foundations that will enable up to date markerless tracking of soft tissues and applies these techniques to analyse breast deformation. Firstly, a novel calibration technique for high-precision camera systems suitable for imaging the surface of the breast is described. The proposed methodology works even when the measurements made by the camera system are more precise than the fabrication methods of the calibration targets. Secondly, to increase the robustness of the breast surface tracking, it describes how sparse features observed in multiple cameras are incorporated within a parameterised mesh surface. This allows for the creation of parameterised time-varying meshes that represent a deforming surface. Finally, a mesh-based refinement scheme is introduced for taking sparse features and densely tracking the surface of the breast. A time varying mesh serves as an estimates of motion, which is then refined to densely track the breast surface. Validation was performed by comparing the proposed markerless approach with a marker based approach. The breast surface was imaged with a multi-camera system designed to track the surface of the breast. During arm movements that caused up to 35 mm displacement of the breast skin surface, the approach recovered breast surface displacements to within 0.5 mm of the marker-based approach. Together, these contributions provide the foundation for the creation of stereo camera systems capable of tracking the deformation of the breast, with the goal of supporting clinicians as they improve outcomes for patients with breast cancer.