These days Artificial Intelligence (AI) is being rapidly developed and utilized in various industries. In the healthcare field, AI is also expected to improve the speed and quality of services, including health risk estimation and prediction, image diagnosis, disease diagnosis, choice of treatment, automatization of surgery, prognosis estimation, and the like. Then it is first indispensable to educate staffs who will be engaged in research and development in medical AI. In Japan, universities having healthcare departments and other institutions are contributing to education of professionals who will work on medical AI in genomic medicine, image diagnosis, drug innovation, and so forth.
For image diagnosis by medical AI, Convolutional Neural Network (CNN) model is widely used by way of transfer learning, particularly Image Classification, Object Detection, and Semantic Segmentation. Its appropriate use can effectively assist in diagnosis and filming as well as reduce overlook of disease. However, appropriate planning is required in advance to decide the purpose and the clear objective of the medical AI, to choose the kind of image, to create supervising image data, to select the model, and the like.
Although artificial intelligence (AI) plays great roles now in various fields, including healthcare, the development of AI has been significantly tough and has undergone different failures and improvements. In the past, AI’s thinking process depended on human thinking and knowledge. At present, AI can solve problems by using the best calculation formula irrespective of them. Non-linear analysis has become possible with multi-layered perceptron. Moreover, the used functions have been improved from the Heaviside to the logistics (sigmoid) and the ramp (ReLU) function, and backpropagation methods and steepest descent methods have also been developed to update parameters.
AI is superior to humans in that it can memorize and dispose of a huge amount of information at the same time, and it can also work extremely in healthcare, such as diagnosis (especially images) and drug discovery (drug repositioning). However, AI’s thinking process remains in a black box, thus AI is expected to work as an assistant to doctors rather than the leading person in healthcare. Particularly, assistance by AI is considered as greatly important in local regions that are short of doctors. In addition, a large amount of information (big data) is required for AI learning, and cooperation between medicine and information processing is indispensable.
Even if artificial intelligence (AI) is used in the healthcare field, doctors will still have the authority and responsibility to decide medical practices and AI will be treated as just a part of medical devices. However, the current laws do not consider significant quality improvement of medical devices by learning. It is required to consider how AI should assist healthcare professionals.
Moreover, big data is indispensable for the improvement of AI, and sharing medical information is being developed to create big data. However, medical information is a huge collection of personal information to be dealt with carefully. Therefore, it must be carefully considered how to protect personal information.
In medical practice, accountability for judgement process is essential, and it is also true in the use of AI for healthcare. Healthcare professionals must be familiar with the structure of AI to some degree because they are held responsible for the use of AI. However, it is extremely difficult to understand the structure of AI or its judgment processes. It may be difficult for AI to be widely used in the healthcare field unless these problems are resolved.
There are two historically contradictory concepts for academics by humans: deduction and induction. Deduction is a method to explain phenomena based on a theoretical rule. Induction is a method to explain phenomena based on past experience. For example, deduction explains the fact based on the heliocentric theory that the sun rises from the east and sets to the west. Induction explains it because the same thing has occurred every day until today. In fact, the same is true for artificial intelligence (AI).
In the past, AI is primarily designed using deduction, and modern AI is mainly designed by induction. Previously, AI had worked according to algorithms that were produced by humans. A typical example is a calculator. The mathematical function of calculator is extremely logical. On the contrary, current AI performs image recognition, including medical diagnostic imaging, with algorithms that are calculated based on past data. No further theoretical rules are required. In other words, the contend of AI has become a black box.
Artificial intelligence (AI) is expected to be used not only in healthcare but also in the fields of health, long-term care, and welfare. The fields of health and medicine are principally made up of prevention, diagnosis, treatment, drug development, and genomic medicine. In the field of preventive medicine, AI may be used for monitoring, support of health life and maintenance, disease onset prevention, and health guidance. In the fields of diagnosis and treatment, AI is expected to be useful for diagnostic imaging support, surgery support, treatment planning, home medical care, telemedicine, rehabilitation, and so forth. In the field of long-term care and welfare, AI is effectively used for caregiving and communication support, and the like.
For AI to be utilized in such fields, medial and health data must first be digitized, unified, and shared. Therefore, a national support system is indispensable. In addition, the development of AI systems is extremely complicated, and requires common understanding among associated industries as well as consideration as to real needs for patients. Moreover, each part of medical device is internationalized, international cooperation is also necessary for AI development.
The use of artificial intelligence (AI) is expected to improve diagnostic imaging support, imaging assistance, the safety, and high-speed imaging in the use of MRI. The main advantages of MRI are non-exposure to radiation, high-quality of inter-tissues contrast, multi-faceted evaluation with various imaging methods, and acquisition of blood vessel images without using contrast media. The main shortcomings are long imaging time, burning risks, and magnetic field attraction. AI is expected to enhance these strengths and to reduce these weaknesses of MRI.
Specifically, AI can facilitate the efficacy of MRI examinations with regard to its patient positioning, imaging method selection, imaging position setting. Immobilization of imaging position is particularly useful to observe changes of the lesion over time. It is also being considered to shorten the imaging time by improving the efficiency of image processing. In interpretation and diagnosis, AI is expected to reduce the workloads to illustrate the lesions and to prevent overlook of the lesions.
In addition, long-time imaging with MRI may cause burns, because its principle is similar to that of a microwave. This adverse effect may be prevented, if such burns can be predicted by performing low-energy irradiation and using AI. MRI is expected to be able to acquire anatomical information, physiological information, and functional information from the subject more efficiently.
It is observed that artificial intelligence (AI) is recently being developed with rapid changes as well as a certain direction of the development. While the number of layers is increasing for deep learning, each component has not become so much complicated. Furthermore, the improvement of performance is becoming more dependent on the scale of big data than the improvement of AI algorithms. Moreover, the development environment is growing, such as libraries, which tends to lower the entry barriers to AI business. The time is coming when anyone can use AI.
However, it is extremely difficult to obtain medical data because of privacy data protection in the field of medical AI. Therefore, it is impossible to enter the medical AI business without obtainment of medical big data, even if the algorithms are publicly available. It is also necessary to hire specialists who are engaged in annotation of medical data for medical AI. At present, it is also studied to train AI with a small number of supervising data by self-supervised learning.