Automatically identifying human disease in 3D magnetic resonance imaging with AI
However, analysing these images is challenging due to the fidelity of anatomical and patient-specific information present in three dimensions (3D). Medical professionals typically do these analyses by hand, which is a time-consuming, mundane task prone to human error and high variability.
Work by the project team over the last decade has recently enabled their algorithms for automatic 3D MR image analysis of human load bearing joints (e.g. knee) for healthy volunteers (with little to no presence of disease) to be productised with Siemens Healthcare, Germany.
However, identifying and accurately highlighting regions of interest (i.e. object segmentation) corresponding to disease and pathologies were not possible at the time, and technologies did not permit having fast runtimes that are highly desirable in the clinic.
Recent work by the authors has provided significant solutions for these problems through the use of state-of-the-art artificial intelligence (AI) methods that are much more accurate and identify pathologies with run-times in the order of seconds as opposed to several minutes.
This project will develop the accurate segmentation and image-based separation of pathological structures corresponding to disease (with respect to healthy tissue) within 3D MR images using highly advanced AI technologies.
Our product will enable large scale studies into diseases such as Osteoarthritis, which is a chronic debilitating disease of many important joints in the human body that has a total economic cost estimated to be US$23 billion/annum globally.
The AI algorithms require resources for the models to be thoroughly generalised and validated on large clinical datasets to ensure robustness, as well as complete the necessary software engineering tasks to form a potential product or grants with global leading partners such as Siemens Healthcare.
Selected Publications by the team:
Dai, W., Woo, B., Liu, S., Marques, M., Tang, F., Crozier, S., Engstrom, C., Chandra, S., 2021. CAN3D: Fast 3D Knee MRI Segmentation via Compact Context Aggregation, in: 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI). Presented at the 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI), pp. 1505–1508. DOI: https://doi.org/10.1109/ISBI48211.2021.9433784
Dai, W., Woo, B., Liu, S., Marques, M., Engstrom, C.B., Greer, P.B., Crozier, S., Dowling, J.A., Chandras, S.S., 2021. CAN3D: Fast 3D Medical Image Segmentation via Compact Context Aggregation. arXiv:2109.05443 [cs, eess] (Accepted Medical Image Analysis) DOI: https://arxiv.org/abs/2109.05443
Chandra, S.S., Lorenzana, M.B., Liu, X., Liu, S., Bollmann, S., Crozier, S., 2021. Deep learning in magnetic resonance image reconstruction. Journal of Medical Imaging and Radiation Oncology 65, 564–577. DOI: https://doi.org/10.1111/1754-9485.13276
Woo, B., Engstrom, C., Fripp, J., Crozier, S., Chandra, S.S., 2022. Anomaly-Aware 3D Segmentation of Knee Magnetic Resonance Images, in: Proceedings of the Fifth Conference on Medical Imaging with Deep Learning. Presented at the Medical Imaging with Deep Learning (MIDL). DOI: https://openreview.net/forum?id=Blt5-qTxdKo