This proposal will develop algorithms and software to improve ultrasound breast lesion detection and recognition. This will be achieved using computer vision to build up a real-time scanning system, alerting the user of the appearance and type of suspicious lesions. The research will improve state-of-the-art methods by providing the location of the lesion, 3D lesions structure and recognition of the lesion.
This proposal is novel as it aims to overcome the problem of inconsistency in the human operator using real-time ultrasound scanning. With the involvement of clinical and commercial partners, we will adopt the user-centred design for real-life usage. We will conduct clinical testing during the lifetime of the project. A low cost, targeted, non-invasive ultrasound breast lesion detection system would have substantial health benefits. This would inevitably result in great long-term societal and economic benefits due to the improved quality of life and health of women in an ageing population.
This improved breast ultrasound diagnostic tool will ensure that healthcare professionals can provide a higher level of care for all patients with breast cancer risks, identifying early signs of breast cancer and referring on to other relevant clinicians.
Aims and objectives
The current performance of computer vision research in breast ultrasound imaging has a number of limitations, including dependence on the human operator, the lack of standardised datasets and algorithms producing high false positive rates. A solution using real time processing methods to overcome these limitations is important and necessary. This proposal will greatly improve the research field by providing new datasets annotated by radiologists and new methods for real-time breast ultrasound lesion detection and recognition. The goal of this research is to provide fast and reliable tools for the early detection malignant lesions. The objectives are:
- (Obj1) Document user requirements and design the data collection tool
- (Obj2) Acquire new ultrasound image sequence datasets with clinical reports, i.e., the location of the lesion and the type of the lesion
- (Obj3) Design and optimise algorithms for real-time lesion detection and segmentation of ultrasound image sequences
- (Obj4) Enhance the lesion recognition technique by using fusion of 2D and 3D features using machine learning algorithms
- (Obj5) Validate the results with clinical decision and conduct clinical testing
For more details visit https://www.mmu.ac.uk/research/research-study/scholarships#ai-65285-10