Challenge: Strokes represent the second most common cause of death worldwide
At UCI, we are using artificial intelligence to detect, quantify, and alert clinicians to the presence of a stroke in real time; providing precious minutes to treat patients.
Stroke Collaborators:
Lydia Su, Ph.D. – Professor, Radiology & Physics
Peter Chang, M.D. – Co-Director for Center for AI in Diagnostic Medicine
Daniel Chow, M.D. – Co-Director for Center for AI in Diagnostic Medicine
Wengui Yu, M.D. Ph.D., Director, UCI Comprehensive Stroke & Cerebrovascular Center
Challenge: People with Dementia
Spatial patterns can inform disease mechanisms
We can map the patterns of cell states back to the region of the brain.
Deep learning can help discover new patterns of cell states and provide new insights into decoding the 3D functional relationships of cells.
Breast cancer:
Challenge:
1 in 8 women will be diagnosed with breast cancer in their lifetime. 1 in 5 breast cancers will be missed by screening mammograms, which are missed opportunities for earlier treatment. 1 in 2 women will be inaccurately labeled with an abnormal mammogram over a 10-year period, which leads to unnecessary tests including invasive surgical biopsies.
We are working on:
- Introducing an AI enabled risk-stratified screening approach that will reduce over-diagnosis while maintaining the benefits of screening – a step towards personalizing medicine.
- Development of AI techniques that can objectively characterize breast tissue inorder to determine risk.
AI automatically segmenting whole breast fibroglandular tissue, a known risk factor for breast cancer
Prostate cancer:
Deep learning enabled evaluation of prostate MRI with automated assessment and identification of nodules at risk for cancer.
Challenge: 1 out of 7 men are diagnosed with prostate cancer. Of 1,000 men offered PSA-based screening, 240 get a positive result, which may indicate prostate cancer. Of those, 100 get a positive biopsy showing definite cancer. 80 choose surgery or radiation treatment.
We are working on:
- There is a lack of non-invasive tools that can differentiate aggressive from non-aggressive cancer types. As a result, only 1 death is avoided for every 48 treated men.
- There is a lack of tools that can distinguish patients who would benefit from existing treatments versus those who would require newer treatment strategies, including targeted therapies
We seek to leverage deep learning with prostate MRIs to optimize detection of high risk and low risk cancers. If successful, this tool can be deployed clinically and be used to guide clinical decision making.
Glioblastoma cancer:
Challenge: Glioblastoma is the most common and aggressive form of brain cancer. 240,000 people worldwide are diagnosed with GBM annually. It is the most deadly form of human cancer with a median survival rate of 14.6 months
We are working on:
GBMs’ underlying genetic variation between and within patients contributes to its poor prognosis.
- We have developed a virtual biopsy application achieved diagnostic accuracy of 94%, identifying relevant genetic mutations.
- Quantitative imaging assessment: our AI application development portfolio includes tools for series identification and an award-winning application for fully automated brain tumor segmentation.
We plan to develop a state-of-the-art deep learning approach to answer these questions:
Can deep learning accurately identify unique GBM subtypes that may have different sensitivities and resistances to therapy?
Can survival in GBM patients be predicted by integrating both imaging and genomic data?
Can we describe the underlying genetic variation within GBMs by leveraging surgical sampling from multiple regions within a tumor?