Medical imaging plays a central role in the diagnosis, treatment and monitoring of cancer – whether it involves visualizing the location of a tumor for surgery or tracking disease progression over time. And artificial intelligence is completely transforming the process, helping to improve diagnostic accuracy and lower health care costs.
Researchers at the UF Health Cancer Center are currently developing AI algorithms for a range of medical image analysis tasks, such as multimodal image fusion, cancer detection, organ segmentation and enhanced image resolution.
“Our goal is to create personalized, effective and less expensive health care solutions that reduce the burden of diseases like cancer,” said Wei Shao, Ph.D., an assistant professor of quantitative health in the UF College of Medicine and a faculty member in UF’s AI Initiative, who is leading this effort in AI-enhanced medical imaging.
Creating solutions
Shao’s lab is called the Medical Imaging Research for Translational Healthcare with Artificial Intelligence Laboratory, or the MIRTH AI Lab. One of the lab’s main projects involves using AI to improve prostate cancer diagnosis. Prostate cancer is the most diagnosed cancer, excluding skin cancer, and the second leading cause of cancer death in men in the United States.
Currently, the most accurate diagnostic test for prostate cancer is a two-step process. A radiologist identifies suspicious lesions on an MRI, and then a urologist performs an MRI-ultrasound fusion-guided biopsy. The method has limitations, including the high cost and the limited number of well-trained radiologists to read scans, Shao said.
“We’re developing a cost-effective alternative to MRI, especially for people in smaller health care systems and rural areas, so they can have a high-quality tool that’s less expensive,” Shao said.
A new technology called micro-ultrasound has a three to four times higher resolution than the conventional ultrasound, allowing urologists to directly target cancerous areas to be biopsied. Studies have shown that the diagnostic accuracy of this technology is similar to that of MRI-guided biopsies.
Applying AI
The main challenge in the promotion of the new micro-ultrasound technology is that urologists generally need additional training to read the images themselves. That is where Shao’s team comes in. The team is developing an AI algorithm to automatically identify cancerous areas on micro-ultrasound, essentially serving as the eyes of the radiologist in real time for the urologist performing the targeted biopsy.
To train AI algorithms to do this, his team has acquired images from 18 patients at UF Health and 72 patients at the University of California, Los Angeles, all of whom had their prostates surgically removed. For those patients, the team has developed a proof-of-concept image registration method to map cancer outlines from histopathology images to micro-ultrasound images.
The second phase of the project involves using aligned radiology and pathology data to develop AI models to automatically find tumors on micro-ultrasound images. This is achieved by extracting meaningful imaging features using deep representation learning. In the final aim, Shao’s team will recruit UF urologists with various levels of experience to assess the new technology’s clinical use.
“Our goal is for this AI model to be as good as an experienced urologist,” Shao said. “This is a step forward to push this technology into clinic use and be adopted by more people.”
Shao, who is also applying AI to prostate cancer screening, earned a master’s degree in mathematics and a Ph.D. in electrical and computer engineering from the University of Iowa, where he developed AI algorithms to improve lung cancer therapy. He completed postdoctoral training in deep learning and medical imaging at Stanford University. And he was recently named a senior member of the Institute of Electrical and Electronics Engineers, a significant milestone that recognizes his contributions to research, leadership and service.
Looking toward the future
Shao is optimistic about how AI can improve health care, noting that increasing amounts of data are needed to improve the accuracy and generalizability of AI models.
“I think the future is bright,” Shao said. “If we can enhance the accuracy and reliability of the algorithms, as well as focus on explaining how models make decisions, we can build trust with clinicians.”
To gather more data, Shao prioritizes collaborations with other universities, including the University of California, Los Angeles and Stanford University. He also collaborates with researchers across UF through the Cancer Center’s Biostatistics and Computational Biology Shared Resource, where he is a member of the Cancer Imaging Research Focus Group.
The key to success in the AI field, Shao noted, is a driving belief in the value of the work.
“I’ve found that most successful students are really motivated and believe we can use AI for good to help people,” Shao said.
