Structured light 3-D surface modelling for the DIET breast cancer screening system

Digital Imaging Elasto-Tomography (DIET) is a non-invasive and radiation-free breast cancer screening method, which is promising for younger women who currently lack effective screening tools. The method applies a mechanical vibration to the breast and measures the motion induced on the surface of the tissue. Surface motion is a function of the internal tissue stiffness. Thus, much stiffer cancer tumours can be identified through the post-analysis of this measured motion. The diagnostic ability of DIET to detect cancers depends upon the accuracy with which motion can be measured. Motion is measured by a computer vision system, and captured images of a breast are post-processed to develop a three-dimensional motion model. Results from early- stage clinical trials highlighted specific limitations of the then current DIET system to model irregular breast surface geometries. This research addresses these limitations. A structured light system projects a pattern of light onto a surface to attain 3-D geometric information. Such a system, consisting of five multi-line lasers was developed to accurately capture irregular breast surface geometries. This thesis focusses on the processing methods of these images to generate a 3-D structured light based surface model of a breast. The processing methods central to the structured light system are laser detection, laser calibration, point-cloud generation, and surface parameterisation. The location of laser beams are identified to a sub-pixel resolution using a two-stage process of laser pixel identification and model-based peak detection. A calibration process estimates a model describing the propagation path of a multi-line laser’s individual beams in space. By projecting laser beams identified in an image onto the calibrated laser model, a breast point-cloud is generated. Finally, a parametric model is fitted to this point-cloud, describing a breast’s surface relative to a polar coordinate system. Each of these methods were integrated into a software solution written in C++. A clinical trial, consisting of a cohort of 21 patients, was carried out to assess the measurement quality of the structured light system. Model agreement to point-cloud data provided a general indication of accuracy with no external measurement system available to attain a true quantification. Visually, it was apparent that the developed methods enabled the DIET system to model geometric irregularities successfully