In the summer of 2016, an innovative medical diagnosis was made at the University of Tokyo's Institute of Medical Science in Japan. Artificial intelligence (AI) that learned about 20 million research papers diagnosed a rare form of leukemia in a female patient in her 60s, which is said to be difficult to diagnose even by specialists, in just 10 minutes. AI helped doctors devise the optimal treatment and saved her life. Since not many cases had been reported on AI's contribution in the medical field, many media outlets took the opportunity to report on the topic.
AI Learns on Its Own to Acquire Knowledge of Medical Specialists
As represented by the report above, AI is beginning to be used as a powerful tool for solving various tasks in the medical field. Why AI? The biggest reason is because environments have been developed where AI can make discoveries and judgments leading to faster and more accurate diagnoses than humans can perform. Some point out that the limits of human cognitive abilities are now becoming a hindrance to the progress of life science research; therefore, an approach has taken shape that "solves various problems in the medical field by harnessing AI to handle tasks it excels at."
AI has two main features.
The first feature is the ability to handle enormous amounts of information instantaneously. Humans cannot process large amounts of information instantaneously, and if processing continues for a long period of time, humans get tired and become less accurate. Even if they could read 20 million research papers, they could not instantly recall any information from the papers.
The second feature is that AI can improve accuracy by learning, which has progressed rapidly with deep learning entering the scene. By analyzing a large amount of data, deep learning learns on its own where to focus in the data to find an answer. This feature of deep learning allows AI to become smarter by itself through learning a huge amount of data, even without proper instructions from humans.
In other words, AI is like a robot that can continue to process a large amount of information without being interrupted, thereby increasing its accuracy along with the amount of data processed. If you provide appropriate data for AI to learn, AI will learn by itself and acquire expertise comparable to that of specialists.
Of course, AI is not almighty. Even AlphaGo, which became famous for having beaten a top-level professional Go player, is nothing more than just a system that can choose the next best step in a Go game; but still, it is a powerful tool when considering the next step to win the Go match. Likewise, if given a specific role in the medical field, AI will strongly support doctors in medical treatment.
AI Is Great at Diagnostic Imaging and Processing Large Amounts of Data
Well, where in the medical field is AI being used? AI is being used to solve problems in current medical practices. The following are some application cases.
Image Interpretation Technique in Diagnostic Imaging
Interpreting computed tomography (CT) images requires professional skills; however, there are not enough radiologists who can read CT images. For this reason, the load tends to concentrate on some specialists. The same can be said of pathologists who interpret pathological images. Since diagnostic imaging is an important first step in starting treatment, any oversight of abnormalities should not be permitted; however, specialists are facing difficult diagnostic challenges in a harsh environment under a concentrated workload. There is also an absence of specialists at many medical institutions. Under these circumstances, there is a need for an AI imaging diagnosis support system that can interpret radiation and pathological images, detect small lesions that even specialists could miss and present suspicious disease candidate names.
To resolve these issues, Fujitsu Laboratories Ltd. and Fujitsu R&D Center Co., Ltd. developed technology for retrieving similar disease cases from a computed tomography (CT) database of previously taken images. The technology, developed with the cooperation of Hiroshima University, works by retrieving similar cases of abnormal shadows expanding in a three-dimensional manner. This technology can accurately retrieve CT images of similar cases in which abnormal shadows have spread in three dimensions by dividing up the organ spatially and recognizing the spread of the abnormal shadows in each area, in the same way as doctors do when determining similarities for diagnosis.
In joint research with Professor Kazuo Awai of the Department of Diagnostic Radiology, Institute and Graduate School of Biomedical Sciences, Hiroshima University, this technology was tested using real-world data, and the result was an accuracy rate of 85%, which is in the top five retrievals among correct answers predetermined by doctors. This technology is expected to lead to increased efficiency in diagnostic tasks for doctors, and could reduce the time required to identify the correct diagnosis for cases where it previously took a great deal of time to identify.
Anticipatory Care for Psychiatric Diseases
In psychiatric medical care, it is not easy to make use of patients' past medical records, which has been recognized as a problem. There is a large volume of medical records, and doctors often describe medical records in a free format using different ways of expression and analysis. Even for patients with the same symptoms, medical record texts are often written in different ways by doctors. Therefore, in many situations, huge amounts of medical record information cannot be utilized in medical examinations.
San Carlos Hospital in Spain used AI to solve this problem. A failure to provide early and appropriate prevention or treatment for a mental illness may result in it turning into a chronic illness or even death. Doctors must therefore make appropriate and quick decisions while properly understanding the health risks to the patient. This requires considering patient symptoms from many sources of information, instead of only clinical examination history. Traditionally, doctors had to read materials printed in various formats. Just selecting the patient records necessary to make a diagnosis took hours.
To support doctors in making quick diagnostic decisions, San Carlos Hospital, Fujitsu Laboratories of Europe, and Fujitsu Spain jointly developed the Advanced Clinical Research Information System based on AI and performed more than six months of field testing. As a result, it successfully reduced diagnosis time per patient by 50%. Five psychiatric specialists at the San Carlos Hospital, each with nearly 20 years of experience, evaluated the field tests. As a result, the system successfully calculated the risks of suicide, alcohol dependency, and drug dependency with an accuracy of 85% or higher.
Comprehensive Clinical Support for Remote and Depopulated Areas
Remote and depopulated areas are suffering from a chronic shortage of doctors. In depopulated areas or remote areas where many elderly patients are suffering from multiple diseases, one doctor must confront various diseases and the ability to detect relevant diseases from a number of symptoms is required. It would be good if general practitioners who received general practice training could be deployed in such areas, but there are not enough general practitioners; therefore, many doctors have to provide general clinical services in harsh environments.
AI has already started to be used to solve this problem. For example, Jichi Medical University and several companies are jointly developing White Jack, an AI-based comprehensive diagnosis support system that displays a list of a patient's possible diseases with probabilities and offers treatment advice for doctors by displaying recommended examinations and prescriptions on the patient's electronic health record. During diagnosis, if a doctor enters physical findings or examination results of the patient, AI updates the analysis and displays a more accurate list of potential diseases. If completed, this system will be a strong supporter for inexperienced young doctors as well as many doctors struggling in remote and depopulated areas.
AI Makes Diagnostic Activities More Familiar and Expands Healthcare Area
How will AI change medical care when AI diagnostic support will spread in the future? A Stanford University research group achieved the following. By using deep learning, the research group successfully diagnosed skin cancer from images with accuracy comparable to a dermatologist. This mechanism, if installed on smartphones as an app, will make it possible to diagnose skin cancer with an accuracy equivalent to a specialist.
Of course, the team needs to do further testing and have specialists check the results before putting it into practical use. However, smartphones are equipped with various sensors, including cameras, and there are also environments to implement AI, so if the data taken with individual smartphones are stored in the cloud and AI continues learning with the stored data, the accuracy will be further improved.
Evolution of AI-assisted diagnosis will make improvements in the medical front by relieving the burden placed on doctors, whereby increasing diagnostic accuracy, reducing patients' stress and innovating the healthcare field through early detection of diseases in daily life. AI-assisted diagnosis is expected to provide support for users' new healthy lifestyles.