Application of deep learning generative network algorithms linked with super-resolution methods and fusion techniques to improve the quality of noisy and unlabeled ultrasound images

We aim at developing new approaches which have not been performed so far, such as i) enhancing the resolution and quality of noisy and unlabeled conventional ultrasound images through deep learning SRMs, ii) synthesizing high-quality MRI from the enhanced ultrasound images through generative networks, iii) fusing ultrasound and synthetic MRI images for different aspects such as classification, early diagnosis, clustering, segmentation, iv) employing few-shot learning approach in the adversarial training process to reduce processing time, computational costs and training data and others, and v) assessing the developed algorithms via quantitative and qualitative validation, different public ultrasound datasets and available ultrasound data. The partner benefits from this project through developing and sharing an open-source code for future medical imaging research, attracting top talent in a competitive AI job market, several scientific publications, processing and analysis of ultrasound images, and gaining domain knowledge through interaction with the intern and intern’s academic/clinical supervisor and others.

Faculty Supervisor:

Ilker Hacihaliloglu

Student:

Partner:

Microsoft Canada Development Centre

Discipline:

Life Sciences

Sector:

Artificial Intelligence; Health and Related Sciences & Technology; Technology

University:

The University of British Columbia

Program: