Currently, many epoch-making attempts by AI are being made in each industry, but it is said that medical care has made dramatic progress. How are changes in medicine made?
Significantly reduce the risk of overlooking cancer by utilizing AI ——The hurdle is delayed Approval of AI medical devices
Artificial intelligence (AI) development in the medical field is in full bloom now.
Efforts to support artificial intelligence diagnosis have begun in various fields such as radiation, pathological images, and clinical examination data.
An endoscopist who reads thousands of images in an hour "I'm scared to overlook"
In September, a medical startup was established to provide a diagnostic system based on artificial intelligence for endoscopic images. Its name is Zubari, "AI Medical Service" (Minami-ku, Saitama City). The driving force behind this business is Dr. Tomohiro Tada, a graduate of the University of Tokyo School of Medicine. He has a 20-year career as a gastrointestinal endoscopist.
Tomohiro Tada, Chairman of the Department of Gastroenterology and Anatomy, who is trying to utilize artificial intelligence for endoscopic diagnosis.
Photo: Takuma Imamura
Japan's endoscopic technology is said to be at the top level in the world. Still, "we are always afraid of overlooking cancer," Tada confesses.
While showing an endoscopic image of early gastric cancer, Mr. Tada asked, "Do you know where the cancer is?" The image showed a reddish stomach wall. However, the amateur did not know at all what the lesion was. No wonder. When Mr. Tada asked the same question on a membership site specializing in doctors, the correct answer rate of about 8,000 doctors who answered was only 31%.
"Of course, endoscopy specialists can tell more, but the number of endoscopy images that a limited number of specialists interpret daily is enormous. In addition to our daily examinations at our hospital, we I am also in charge of "double check" to verify the examination images of the local government, but when there are many, I sometimes read more than 3000 sheets in one hour. To be honest, it is impossible for humans to continue this work for more than one hour. I can't keep my concentration. "
"The Urawa Medical Association, to which I belong, has more than 2 million images of gastrointestinal endoscopy and screening per year in double-check work. I have been wondering if this work can be simplified. I did. "
Mr. Tada, who felt such a dilemma in the field, heard from Mr. Yutaka Matsuo, an artificial intelligence scholar, that "AI image recognition exceeded human ability", and "humans and artificial intelligence double-check. I want to create a system that can do it and eliminate oversight of cancer! ", He said, thinking of using AI to assist humans.
Mr. Tada started the development of an artificial intelligence diagnosis support system for gastroscopy in November last year. In collaboration with a major hospital, a patent has already been applied for a diagnostic system for Helicobacter pylori gastritis, which is the cause of gastric cancer. When we verified the accuracy of the test, we found that "although it is not as good as the top doctors, it has already achieved performance that exceeds the average of doctors." In addition, "real-time diagnosis" using video has entered the stage where it is possible.
"If these systems permeate the medical field, specialists can closely check the lesions picked up by AI to improve the accuracy of examinations, or the assistance of artificial intelligence will reduce doctors' careless mistakes. So to speak, car safety. It's like a back monitor that supports driving. If we can reduce oversight of cancer by enhancing the capabilities of doctors, I think that doctors and patients will become “win-win” with each other. ”(Mr. Tada)
24% missed due to technical disparity among doctors
As Mr. Tada points out, the "wall" that hinders early detection and treatment of cancer is the problem of "oversight" of lesions by screening. In fact, what is the risk of overlooking? Let's take the example of colorectal polyps and colorectal cancer.
There are overseas research reports investigating the oversight rate of adenomatous polyps called "precancerous conditions". It is said that 24% were overlooked due to lesions that are difficult to distinguish with the naked eye, the site of occurrence, and the technical disparity of doctors. In another study, 6% of cases later developed colorectal cancer despite having undergone colonoscopy. It is reported that 58% of the causes are "missing during endoscopy".
The AI-based "real-time endoscopic diagnosis support system" announced by the National Cancer Center (Chuo-ku, Tokyo) on July 10 may be a promising option for preventing colorectal cancer from being overlooked. .. First of all, please watch the video.
In this way, AI instantly detects the lesion during endoscopy, and even identifies the location and points out that it is here. In order to accurately capture the lesion, NEC's "face recognition technology" and "pathological image analysis" technology, which is also a joint development destination, is applied. Deep learning has made it possible to distinguish between lesions and non-lesions.
AI learned 130,000 in addition to the lesion images of about 5,000 cases diagnosed by endoscopy at the National Cancer Research Center Central Hospital (polyps as precancerous lesions and images of early colorectal cancer). There are as many as 5,000 "normal images without lesions". When verified using about 5,000 data for accuracy evaluation separately from the images for learning, the correct diagnosis rate of cancer and polyps of "precancerous state" was about 98.8% (for those with lesions, " It was found that "yes" and "no" can be distinguished by the probability of correctly diagnosing "no".
About 1600 cases of "flat / depressed lesions", which are considered to be relatively rare, are also learned. In the case of this type, there are cases where the lesion infiltrates deep into the tissues of the body at an early stage, and it is found only after the cancer has become invasive, which is difficult for doctors to recognize with the naked eye, but for development. According to Dr. Masayoshi Yamada of the Department of Endoscopy, National Cancer Center Hospital, who was involved, "AI can be distinguished at the same level as humans even in the" flat / recessed type "that is difficult to distinguish unless you are really an expert. It came to be. "
Mr. Yamada points out that there are two main merits of introducing AI into endoscopic diagnosis. One is that AI support reduces human mistakes.
"Even though we are doing endoscopy, we sometimes get impatient in situations where we have to hurry, and there is always emotional unevenness. But they (AI) are so calm that they are always the same. "There is a lesion here."
The other point is to eliminate the technical gap between doctors.
"Of course, the final decision is humans, and AI does not support humans, but the discovery rate is such that" as long as there is a lesion there, anyone can catch it. " Personally, even if a doctor who is doing medical treatment alone on a remote island inspects it, or if it is inspected at a large hospital in the city, the accuracy of the inspection is the same, so it can be used anywhere in the country. I want to make it a system "
However, there is a "wall" for implementation.
The National Cancer Center is aiming to start clinical trials two years later, but the implementation will be "depending on the timing of the regulatory approval review" (Mr. Yamada). In fact, the current situation is that Japan does not even have approval standards for artificial intelligence medical devices. Mr. Tada mentioned above is also raising a sense of crisis regarding this wall toward implementation.
"Although endoscopy is a leading field in Japan, the development of inspection systems for the utilization of AI is an area where the world is competing. Even if there are many products that have been developed to the point where they can be used almost, I hope that Japan will not be left behind in the world while no one gets approval and it cannot be used in clinical settings. "
What are the challenges of "breast-only PET" that can see up to 5 mm cancer?
On the other hand, technological innovation has progressed in various fields other than medical devices that use artificial intelligence. For example, a breast cancer testing device.
However, in breast cancer screening, the problem of "overlooking" as well as "overlooking" emerged. In Japan, it has swayed between these two walls.
Many Japanese women have "high-density breasts" that are difficult to detect abnormalities only by mammography, which the country recommends to women over 40 years old. The more developed the mammary glands, the whiter the X-ray image will look, and the cancer may be hidden behind a white shadow.
In October last year, 32 breast cancer patient groups nationwide submitted a request to the Ministry of Health, Labor and Welfare, saying, "Even if you take the test, it should be overlooked ...". As a result of the examination, he requested that he be informed if it was difficult to determine the presence or absence of cancer in high-density breasts.
Higher density breasts to the right of the image. The mammary glands appear white, making it difficult to diagnose cancer.
Photo: Provided by NPO Breast Cancer Imaging Network
Ultrasonography is a way to detect cancer that could not be read by mammography. Studies have shown that when used in combination with mammography, the rate of breast cancer detection was 1.5 times higher than that of mammography alone.
However, as a result, there were still issues such as an increase in the number of "false positives" who were suspected of having breast cancer even though they did not have breast cancer. You may experience unnecessarily emotional distress until you confirm that you have cancer. At this time, the country does not recommend echo as a countermeasure screening. Those who wish to undergo a medical examination are basically responsible for the full amount.
As a new option for breast cancer testing, Shimadzu announced in September an improved version of the breast-only PET device "El Mammo."
"PET (Positron Emission Tomography)" is a test that utilizes the property that cancer cells ingest 5 to 8 times as much sugar as normal cells. Instead of looking at the "morphology" that is copied like mammography or ultrasound, a drug that closely resembles glucose (a drug called FDG that emits radiation from glucose) is intravenously injected, and cancer cells take up the sugar. It is a method of scanning the place that is metabolized in, that is, "what you ate". The principle of PET examination is the same for the whole body and for the breast. In the case of "breast only", a detector is placed around the hole where the breast is inserted, achieving higher sensitivity and resolution than whole-body PET. I can see even a small cancer of about 5 mm. Cases have been reported in which lesions could be detected even in high-density breasts, which are difficult to detect with mammography, using this device.
However, even if this breast-only PET device can overcome the "invisible wall" peculiar to mammography, there is "another wall".
First of all, the cost is high and the number of facilities available is small. Breast-only PET scans are not covered by insurance for preventative screening. Generally, it is performed as an option for whole-body PET cancer screening, so the out-of-pocket cost exceeds 100,000 yen. Insurance is applied to patients suspected of having cancer, but in Japan, it is not recognized as insurance medical treatment unless it is a combined examination with whole body PET, so it costs a double examination cost for whole body and breast only. Will end up. Therefore, the burden on the patient is about 40,000 yen when the burden is 30%. The number of facilities available in Japan is currently limited to about 10.
Furthermore, even in inflammation such as "mastitis" and benign breast tumors called "fibroadenoma", some of the drugs have the property of collecting and are projected as lesions. In that case, it is difficult to distinguish between good quality and malignant. As with ultrasonography, there are challenges in detecting false positives.
An improved version of the breast PET device "El Mammo". Achieves a "painless examination" that does not put pressure on the breast and reduces the burden on the person undergoing the examination.
Provided by: Shimadzu Corporation
Munetaka Takahashi, group leader of the product at Shimadzu Medical Equipment Division, said:
"It's a testing device that can detect the presence of cancer at a fairly early stage, but like other testing methods, it can't cover all types of breast cancer. One possibility is testing. In the future, there is a direction to save high-risk people such as familial and hereditary breast cancer who could not be identified well by other test methods by taking advantage of the high accuracy of I think"