Mike Barlow has this view in “Artificial Intelligence and Medicine”:
Unlike traditional healthcare, which tends to be labor-intensive, emerging healthcare models are knowledge-driven and data-intensive. Many new types of healthcare are bringing us a new model that will rely on a new generation of user-friendly, real-time big data analysis and artificial intelligence and machine learning tools.
There are many experts with similar views as Mike Barlow. Medical care has become one of the first scenes of artificial intelligence. It is no longer a technical problem for AI to enter the hospital from the laboratory. It is just a matter of time.
There seem to be some exceptions in reality. As early as 2014, IBM invested $1 billion to establish the Watson Business Group to fully operate Watson’s research and commercialization issues. The first job was “doctor assistant”, trying to use artificial intelligence to drive personalized treatment.
But to this day, case reports about AI+medicine are not uncommon, but the clinical implementation of medical AI is almost in a blank state.
01 Character Evolution
Since 2014, AI has entered the field of vertical segmentation, and medical + AI has been considered one of the easiest areas to land. Although there are still many problems in medical + AI, the pace of progress has never stopped.
Watson, owned by IBM, is a pioneer in this industry. In September 2017, Watson achieved its landing in China, and its role is positioned as an “assistant doctor.” In other words, Watson can’t make judgments on behalf of doctors, but uses artificial intelligence to help doctors make diagnoses. Just like a golden hoop, Monkey King has it even more powerful.
The working principle of Watson is that when the doctor enters the detailed data of the patient, this AI will search the published research results from the database. In less than 10 seconds, the corresponding treatment plan can be given for the doctor’s reference and advice for the doctor. Program.
Although Watson’s current role is to assist, but in order to do a good job in the role of “assist”, Watson is constantly self-optimizing.
Take Watson’s oncology solution as an example: currently its program covers 13 cancer types including breast cancer, lung cancer and rectal cancer, assisting doctors around the world in diagnosis and treatment, and it is expected that by 2019, three more will be added to the existing basis Solutions for cancer species. In China, domestic Internet giants have also entered the AI medical field.
In April 2016, Tencent invested US$100 million in “Carbon Cloud Intelligence”. In October, Baidu released the “Baidu Medical Brain”. In March 2017, Alibaba Cloud released the “ET Medical Brain”, announcing its official entry into the AI medical field.
Although it started two years later than the United States and its technology is not as mature as abroad, it still has certain barriers in the domestic market.
1. The physical fitness of Chinese people is different from that of foreigners. The AI medical system injected into Watson and other foreign countries is created for the physical condition of foreigners and may not match the Chinese.
2. According to laws and regulations, domestic medical data is not allowed to leave the country. Therefore, the AI medical system that serves the Chinese needs to be completed by the Chinese themselves.
02 Spiral
In recent years, the concept of AI has been hyped up. Some people even say that as long as the word “AI” is added to the PPT, it will definitely attract investment.
So as a medical AI, is it a “mirage” or a revolution that is truly subverting the medical industry?
1. “It’s just a pile of shit”?
According to the US media STAT, IBM’s internal PPT shows that Watson actually has serious technical problems. IBM’s medical experts and customers have confirmed multiple unsafe and incorrect treatment recommendations.
Watson’s technology did not meet expectations, causing complaints from doctors all over the world. Many doctors said Watson is not suitable for patients in their own countries.
Some doctors think that Watson is not of much use, and that the hospital purchased Watson for marketing purposes. There was even a doctor who told IBM bluntly, “This product is just a shit, and it’s useless most of the time.”
A lady once mocked the use of Faraday’s “electromagnetic induction device”. Faraday replied, “It won’t take long to collect taxes.” Since then, how much electricity has changed the world, I believe everyone understands.
It is true that medical AI at this stage is not very intelligent, nor can it completely replace doctors. But we have to see the advantages of AI. For example, the human brain is fragile, while the machine can perform endless high-intensity calculations.
From the perspective of development, consider AI medical care. What this new thing can do may be completely beyond people’s imagination.
2. Suspected of over-promotion?
Since the news that AlphaGo defeated human chess players, the topic of AI has been the darling of the media. This overheated topic may have given the public too high expectations for artificial intelligence at this stage.
So, for example, when Watson layoffs 50% to 60% of its staff, for example, the Sun Yat-sen University Eye Center found that the accuracy rate of AI doctors in real clinical outpatient diagnosis of cataract was only 87.4%, which was far lower than the 98.87% in the trial phase. You can’t help but question-Has medical AI failed?
In fact, this is just a situation that things often encounter in the process of spiraling upward. The problem is that the excessive publicity of the media prevents the public from having an objective perception of things. The public has over-imagined technology, deified technology and simplified difficulties.
Medical AI has its own spiral upward cycle, the public should allow failure, and the media’s attention to this issue should return to rationality.
03 remaining issues
The birth of new things is always accompanied by pain. Medical AI needs to overcome its own shortcomings.
At the same time, the birth of new technologies often suffers from the old order, and medical AI also needs to face the problem of how to break through the old order.
1. Legal Liability
In the medical field where the cost of trial and error is always life, AI is the first to face legal issues.
In 2011, in a hospital in Massachusetts, an elderly man who collapsed was taken to the emergency ward by an ambulance. He was immediately installed with AI physical sign monitoring equipment. If his vital signs are in danger, the device will issue a warning and call a nurse. However, the next day, the old man died in the hospital bed.
Before his death, the red light of the monitoring equipment flashed all night, but was pressed down by the nurse on duty over and over again. The negligence of nurses is naturally to blame, but from a system perspective, there is a problem that everyone can’t avoid: many hospitals’ AI monitoring equipment is often just false alarms.
For the death of the patient, who should bear the responsibility, and whether AI can take responsibility for the misdiagnosis is a question that needs to be considered.
2. Data island
Just like a car needs gasoline to drive, data is the basis for AI to run. AI increases its “experience” by “eating” massive amounts of medical data, thereby making itself more “intelligent”. However, in China, medical data seems to be abundant, but in fact the availability is not high.
For example, the data exchange between hospitals is not well done. If a patient sees a doctor in different hospitals, it becomes very difficult to obtain complete historical data of the patient.
Moreover, different hospitals use different hardware instruments, resulting in different data formats, making it difficult to standardize. The data between the various hospitals are like isolated islands in the ocean, independent of each other, cannot be connected to each other, and cannot communicate with each other.
Many industry experts called for the conversion of the private format of each hospital data into a standard format so that medical data can be used universally, but there are few respondents.
Even if AI can obtain high-quality medical data, it still has a thorny question that cannot be avoided: Will patient data be leaked by AI companies? After all, no one should want their privacy to be leaked.
3. Hard to land
In addition to data issues, AI’s implementation in the medical industry also has model and system issues.
For example, the Trajanova laboratory in the United States has developed a set of cardiac imaging programs that combine images and AI to construct the entire 3D holographic heart model. It can simulate the dynamics of the heart, using it, doctors can accurately locate the patient’s focus.
However, this technology is really going from the laboratory to the hospital, and the prospect is not optimistic.
The biggest challenge comes from the supervision and review of the US Food and Drug Administration (FDA). If any technology wants to be put into clinical application, it will inevitably engage in a protracted tug of war with the FDA. If the research results cannot be converted into approval standards, no matter how much research results are produced, it is useless.
04 Summary
Although medical AI has not yet been implemented on a large scale, there is still a long way to go from the laboratory to the hospital, and at this stage, medical AI cannot perform diagnosis like human doctors, and cannot replace human doctors.
However, AI medical treatment is a very good tool. Its emergence has effectively improved the diagnosis efficiency of doctors, improved the quality of medical treatment, and reduced the possibility of misdiagnosis.
With the rapid development of science today, we need higher-level and more scientific technology to enter the medical field, and AI may be the best and most appropriate technology of the times. In the future, AI will play a pivotal role in the medical field.
