Detection based on screening programs in high-risk populations is a major asset in the struggle against Breast Cancer. Currently, screening programs in most developed countries focus in analysing digital mammography images. Computer Aided Diagnosis (CAD) systems provide decisive help in this task. Nevertheless, latest tendencies imply that not all women should follow the same screening protocol (i.e. systematic mammography every two years), but they should be stratified according to several criteria. This criteria stand mainly for cancer risk biomarkers such as breast density or previous lesion evolution. The capacity to stratify patients improves diagnostic efficiency by using different medical imaging modalities. Specifically, personalised screening programs make use of the image modality that provide the better information for each patient type.
In order for these biomarkers to reach their full potential in clinical practice, their implementation should allow for fully automatic computation in large patient populations. Additionally, they should also allow the evaluation and interpretation of the specific values for every patient. Finally, they should make the most of all available image modalities. Computer Vision techniques already present via diagnostic equipment represent a very important improvement in everyday clinical practice. Nevertheless, further improvement is still needed and the implementation of new algorithms will make a difference in the coming years. This demand, arises both from epidemiologists and radiologists and satisfying it will play a key role in the achievement of personalised screening programs. Automatic generation of Biomarkers is the next major milestone in this direction.
The IA-BioBreast project aims at researching image analysis methods that focus on the development of two specific biomarkers: breast density and temporal evolution of existing lesions. In order to achieve this goal, new microtexture-specific techniques will be developed using algorithms for feature extraction, selection and classification. On the other hand, image registration algorithms will be researched for two main applications: combining images of different modalities (breast X-ray mammography, MRI and UltraSound) and registering temporal studies within the same image modality. Finally, automatic lesion detection algorithms will also be developed by using image segmentation techniques. The adequacy of the project results to clinical practice will be analysed by using a CAD (Computer Aided Diagnosis) system able to process the results from all the techniques developed. By including all these aspects, the project focuses in using computer vision techniques and developing novel algorithms for segmentation, feature extraction and selection, classification and registration.
The project will benefit from the involvement of several health centres as well as the interest shown by CAD developing companies. Not only these partners will provide data during the development of the project but they will also take an active role in it. Moreover, they will also make it possible to evaluate the methods at a higher, clinical level once they reach completion stages.
