Advances in Breast Cancer Imaging: Three Things Entrepreneurs Should Know
Damon Diehl, PhD, Technology Program Manager
“Nobody knows what the AIs are thinking,” Dr. Zhimin Huo said. “They are basically black boxes.”
It sounded like the beginning of a dystopian novel. Artificial Intelligence (AI) was not the conversation starter I had expected when Dr. Sujatha Ramanujan and I sat down to discuss breast cancer with her former colleague, Dr. Huo.
Dr. Huo is the former Lead Research Scientist at Carestream Health Inc., a medical imaging company that was spun out from the Eastman Kodak Health Group in 2007. She is recognized as a world leader in computer-aided diagnosis and the physics of medical imaging, her research being cited thousands of times in academic literature. It’s important to note, though, that in addition to her immense technical background, Zhimin was also deeply engaged in commercializing that technology.
The recovery rate for breast cancer has been steadily improving since the 1990s. Dr. Huo, Sujatha, and I met to discuss what part optics, specifically imaging, played in that trend. With three physicists sitting at the (virtual) table, the conversation frequently descended into mathematics. Still, during our conversation, I learned some unexpected things that anyone entering this market needs to understand.
AI is driving the advancements. Since 2012, the precision of 3D imaging of breast cancer tumors has been increasing exponentially, and it has very little to do with better imaging technology. These advances are driven by “Moore’s Law.” Increasingly powerful graphic processing units (GPUs) operating in parallel have engendered a new generation of AIs based on “deep learning” neural networks that can self-train, without prior knowledge, by analyzing millions of images.
Better treatments, not better diagnosis. Higher precision 3D imaging’s real power is that these models enable doctors to craft unique treatments for each patient that can target tumors within the breast with very little damage to the surrounding tissue. These targeted treatments minimize exposure to chemicals and radiation, which, in turn, reduces the discomfort patients endure during treatment and also lowers the risk of long-term side effects.
Focus on workflow. The resolution improvements enabled by advanced AIs far outstrip advances in hardware precision. At present, there is no market pull for “better” medical instruments if all they have to offer is an improvement in speed or precision. Instead, an entrepreneur should analyze ways to improve the workflow of detection and treatment. Innovations that eliminate steps in the workflow are far more viable than equipment that replaces or adds to the workflow.
The above three points can be distilled into one deeper observation that every entrepreneur needs to remember: Just because you understand your technology does not mean you understand your market.
I can also provide another observation: We need not fear the deep-learning AIs. They are on our side… for now…