Artificial Intelligence in Medical Imaging: Current Successes and Future Focus
On Sunday, 2 February, as part of 2020 SPIE Photonics West, Kyle Myers, the director of the division of imaging, diagnostics, and software reliability in the FDA's Center for Devices and Radiological Health's Office of Science and Engineering Laboratories, facilitated an industry panel on artificial intelligence in medical imaging.
Presenters Maryellen Giger, a professor of radiology/medical physics at the University of Chicago, and co-founder Quantitative Insights, Charles A. Taylor, founder and CTO of HeartFlow, and Santosh Bhavani, senior director of business development at JLK Inspection, were joined for a post-presentation discussion by Moshe Safran, CEO of the RSIP Vision USA offices.
Giger, whose Quantitative Insights' product QuantX was the first FDA-cleared machine-learning-driven system to aid radiologists in cancer diagnosis, presented on AI-aided breast cancer diagnosis, from lab to product. She detailed her multi-decade-long research at the University of Chicago in developing machine-learning methods for use by radiologists as an aid in diagnosing breast lesions imaged on MRI. She also described the translation of her lab's developments to a clinical workstation through the creation of startup Quantitative Insights which recently sold to Paragon Biosciences as Qlarity Imaging. AI, she noted, is changing healthcare all along the diagnosis-treatment pipeline, and that efficient use of computer can also improve efficiency of humans. "It's important," she added, "to remember that AI is not magic; it depends very much on high image quality."
Taylor, whose HeartFlow utilizes AI for the diagnosis and treatments of heart disease, noted the escalating costs of Coronary Artery Disease (CAD) as well as issues with traditional, invasive techniques of monitoring, assessing, and management of patients with CAD. His HeartFlow technology, which began with an initial question of how to make these procedures less invasive, utilizes data from a CT scan to a 3D digital model of the patient's heart. Using deep learning, it can then pinpoint arterial blockages and predict the impact on blood flow. He considers AI as a means to an end, a tool that his team uses with their approach and products, but affirmed that humans are still "in the loop," with a rigorous human inspection and correction process in place.
Santosh Bhavani, director of business development of JLK Inspection, highlighted multiple JLK products which employ AI in conjunction with brain imaging: ASTROSCAN for dementia, UNISTRO for brain stroke; and JBA for cerebral aneurism. In addition, he described HandMed, a portable handheld AI small X-ray camera can take a chest X-ray anywhere, anytime, and use AI based analysis to generate a report on whether the person has signs of tuberculosis. Other JLK devices include FUNDASCAN, a portable, handheld small camera which offers flexible, accessible analysis for such geriatric ophthalmologic diseases as glaucoma, cataract, and retinal disease; UNIENDO, an AI-based colon polyp and gastric cancer detector; and MammoAna, an AI-based breast cancer detector and prevention system.
Following the presentations, Safran, whose RSIP Vision is a leader in AI, computer vision, deep learning, algorithm development, and image processing technology for the medical device, pharmaceutical and automotive industries, joined the others as they debated such AI-related questions as the importance of managing data properly, the critical need for high quality data, the direction and progress of autonomous learning (still very much in the development phase, seemed to be the general consensus), as well as the importance of providing radiologists with the best possible tools to raise their work to a higher, more efficient level. They also noted the importance of a properly prepared workforce, emphasizing an interdisciplinary approach that covers AI, domain expertise, and biomedicine.