Hailed as giving birth to the "fourth industrial revolution," artificial intelligence (AI) is being developed in yet another field -- cancer diagnosis. Medical and research institutions are developing AI technology to diagnose cancer using "deep learning."
One example is the Japan Society of Pathology, which began developing AI technology in February at 29 institutions, including The University of Tokyo Hospital and Kyushu University Hospital, to reduce the load on the country's overworked pathologists. The system being developed by the society compiles images of tissue samples from patients at the participating institutions into a database, where AI then decides if the case is cancer or not, learning from its past successes and failures, a method called deep learning.
Pathology is a field of medicine where doctors examine samples of body tissues using a microscope and decide whether the patient has a disease, such as cancer. The task of the pathologist to diagnose cancer is an important one. With the number of cancer patients and studies examining the effectiveness of cancer drugs increasing, the number of pathological examinations has doubled between 2005 and 2015.
However, in the roughly 700 hospitals nationwide with over 400 beds, one-third of hospitals do not regularly staff pathologists, and of those who do, nearly half have only one pathologist. That means that the amount of work falling on individual pathologists is increasing, and any mistakes put patients' lives at risk.
One of the doctors involved in the development of the pathological society's AI, Dr. Katsuhiko Saito, is the head of the pathology department at Toyama City Hospital. He is the only pathologist at the hospital. When cancer reappeared in a patient, he had to search for the previous sample among the shelves of a separate room to diagnose the patient. He wondered if the process could be made more efficient. In 2006, he digitized images of the year's 4,000 samples and developed a system for searching the database of photos. He thought if AI could make a preliminary comparison and diagnosis, then the burden could be reduced even further.
The society's AI and database is being developed with this goal in mind. So far, when testing the ability of AI to distinguish between stomach cancer and a similar-looking benign symptom, AI correctly diagnosed the cancer 70 percent of the time. However, it may take another five to 10 years to develop the system enough for clinical use by pathologists.
Another research team employing deep learning in their clinical research on cancer is led by Satoru Miyano, professor at the Institute of Medical Science of the University of Tokyo, using the AI "Watson" developed by IBM.
Research on cancer at the genetic level is quickly progressing, and according to Miyano, roughly 200,000 papers on the subject were presented just last year alone. While it is impossible for a human to read and learn from that volume of research, it is possible for AI.
One success story that gathered attention was that of a woman in her 60s whose leukemia treatment was not going well. AI came up with the name of her disease and the medicine used to treat it in around 10 minutes. "Artificial intelligence is a necessity to assess the condition of cancer that people cannot accurately diagnose. We hope to raise the degree of accuracy of AI through further deep learning to improve treatment," Miyano said.
But the idea of relying on AI in the critical work of cancer diagnosis is also raising concerns. The Ministry of Health, Labor and Welfare created an advisory panel in January 2017 to weigh questions concerning responsibility surrounding treatment using AI, clarification of who makes treatment decisions, and the need for authorization of AI as medical equipment. According to AI specialist and panel member and specially-appointed associate professor Yutaka Matsuo of the University of Tokyo, the biggest difference between robots of the past and current AI is the ability to learn on its own. "Doctors enhance their abilities by accumulating a variety of experiences. AI does the same thing by building up knowledge through tests and actual performance. I think AI will eventually increase its level of trustworthiness," he said.
However, Matsuo cautions against having too high expectations. "Even as AI continues to be developed, there will always be a need for human doctors," he points out. "A doctor looks at body condition as well as outside factors such as occupation and examines the patient as a whole. It is thought that the aim of AI will not be to reach the same level as doctors, but for it to generally take on a role that reduces the burden on doctors."