Research on applying artificial intelligence = AI to medical treatment is underway all over the world, and one of the most advanced fields is diagnostic imaging, which allows AI to learn image data of sick patients and utilize it in the judgment of doctors.
In particular, in ophthalmology, which often diagnoses based on eye images, there is a movement to utilize AI for early detection of diseases and to prevent oversight of diseases.
Four years ago, a group of associate professors Hidenori Takahashi of Jichi Medical University developed a system that instantly checks for suspicion of illness from photographs taken by examining the fundus in the back of the eye using AI.
The system can determine whether there are 100 types of illnesses, and it is used for the diagnosis of doctors, so when it was introduced in the health examination conducted at the hospital, the time required for diagnosis was reduced to about one-third. It means that the number of oversights of illness has decreased.
In addition, Dr. Hitoshi Tabuchi's group, a professor of the Department of Endowment, Hiroshima University, automatically determines from the photographs of the fundus whether there is a risk of "diabetic retinopathy," which can lead to blindness due to damage to the retina due to diabetes. We are in the process of developing and are trying to detect diabetic retinopathy at an early stage by investigating diabetic patients who visit the internal medicine department using AI.
Although there is still a problem with the accuracy of AI's correct judgment, the movement to utilize AI in the medical field is being promoted in Japan as well, as clinical research aimed at practical application is underway at related hospitals.
AI system shows the possibility of 100 kinds of diseases such as glaucoma
A research group of Associate Professor Hidenori Takahashi of Jichi Medical University has AI learn the pictures taken by fundus examination to see if there are any abnormalities in the retina that senses brightness, color, shape, etc. in the back of the eye. We have developed a system to check for suspicions and are using it in health examinations conducted at hospitals.
The research group trained AI to learn the pictures of fundus examinations of about 500,000 patients stored in a university hospital for 10 years and the results actually diagnosed by an ophthalmologist, and whether there is a suspicion of eye disease. I developed a system to investigate four years ago.
In the system, when the picture taken by the fundus examination is read, AI identifies whether there is bleeding in the retina or abnormalities in the blood vessels and nerves in the back of the eye, and what kind of disease is possible. Indicates.
The target diseases are "glaucoma", which may cause blindness due to the narrowing of the visual field, "cataract", which causes the crystal body that acts as a lens in the eye to become cloudy and impairs eyesight, and the retina due to diabetes. There are 100 types of eye diseases such as "diabetic retinopathy" that can lead to injury and blindness, to systemic diseases such as high blood pressure and arteriosclerosis.
The probability that the possibility of the disease indicated by AI matches the diagnosis of the actual doctor is more than 80%, and as a result of introducing it at the "medical examination center" of the university hospital, the time required for the diagnosis of the doctor is one third. It has been shortened to a certain extent, and the number of oversights of illness has decreased.
A system that uses AI to diagnose eye diseases has been put into practical use in the United States to detect retinopathy caused by diabetes, but Associate Professor Takahashi is the first in the world to be able to determine 100 types of diseases at once. It is said that.
Associate Professor Takahashi said, "I used to spend about an hour looking at pictures of the fundus of about 30 people, so the burden was lightened. Sometimes I couldn't believe the name of the disease indicated by AI, but I often look at the images. I sometimes noticed that I had missed a small bleeding. "
Since the system has not been approved by the government as a medical device, it can only be used for health examinations conducted at the patient's own expense, so Associate Professor Takahashi aims to commercialize it so that it can be widely used in collaboration with medical device manufacturers. I am.
Associate Professor Takahashi said, "The accuracy of AI diagnosis is higher than that of trainees for general eye diseases, but it is still insufficient for rare diseases, so I would like to raise it to the level of an experienced ophthalmologist. I would like to make it available for health examinations and general clinics. "In the future, I would like to be able to remotely diagnose patients even in areas where there are few doctors who can diagnose by looking at images.
A woman who goes to the hospital for glaucoma "It is very good to point out early with AI"
The Ophthalmology Department of Jichi Medical University has been using the AI system to check for eye diseases for people who have undergone a fundus examination at the medical examination center of the University Hospital since June.
The number of patients examined per day is about 30 people, and the possibility of illness is pointed out by the AI system, and after that, there are multiple patients who have started treatment after being diagnosed by a doctor and finding out that they have illness. is.
Of these, a 58-year-old woman who was found to have glaucoma and went to the hospital was able to use eye drops daily to prevent her symptoms from getting worse.
She said, "I don't have any subjective symptoms such as eye pain or narrowing of my vision, but if AI can point out the possibility of illness early, I'd like to see a specialist doctor, so the introduction is very good. I think it's good, "she said.
Besides ophthalmology
There is also a movement to lead to early detection of illness by collaborating between hospital departments and performing eye image diagnosis by AI other than ophthalmology.
A group of Dr. Hitoshi Tabuchi, who is a professor of the endowed course at Hiroshima University, automatically determines from the photographs of the fundus examination whether there is a risk of "diabetic retinopathy", which may lead to blindness due to damage to the retina due to diabetes. Has been developing.
Although diabetic retinopathy often progresses without subjective symptoms, treatment is often delayed, but according to a survey by the National Center for Global Health and Medicine, 60% of diabetic patients have not undergone eye examination. Dr. Tabuchi is trying to help diabetic patients to detect diabetic retinopathy at an early stage by using AI to check for the risk of diabetic retinopathy when they visit the internal medicine department.
However, there is a problem with accuracy, and when 70 photographs of the fundus of diabetic patients were applied to AI, more than half of the photographs judged to have diabetic retinopathy match those judged by a specialist ophthalmologist to have this disease. However, some of them were judged to be ill or overlooked even though they did not have diabetic retinopathy.
Although it is difficult to make a judgment based on AI alone, Dr. Tabuchi said that it is effective to some extent to find out who may have diabetic retinopathy among those who have not consulted an ophthalmologist. There is no doubt that it will be, so I think it will be easy for patients to accept. "
In addition, the group is also developing a system in which AI warns that there is a simple mistake such as making a mistake in the left and right eyes during surgery. It is said that the role it can play is great.
Dr. Tabuchi said, "Usually, humans can make mistakes in extremely simple things that no one can make mistakes, so the best way to use them is to have AI support that part. I think. "
AI-related announcements at the "Japanese Ophthalmological Society" one after another
The "Japanese Ophthalmological Society" was held in Osaka to announce the research results related to the medical treatment of eye diseases, and the latest diagnostic technology using AI = artificial intelligence was introduced.
At the general meeting of the Japanese Society of Ophthalmology, which was held from 14th to 17th this month at the Osaka International Convention Center in Kita-ku, Osaka, about 8,000 doctors and researchers from all over the country working on the diagnosis and treatment of eye diseases. Participated and announced the latest research results.
One of the research fields that attracted attention is AI = diagnostic technology using artificial intelligence. At the symposium held on the 14th, Associate Professor Hidenori Takahashi of Jichi Medical University used AI to image eye examinations. We have released the latest data on research to determine the presence or absence of suspicion of illness and to estimate eyesight.
At the symposium, many researchers were paying attention to the presentation on how to create a mechanism to utilize AI in society, such as points to note when using AI for diagnosing diseases.