Shulou(Shulou.com)06/03 Report--

Https://www.cn-healthcare.com/articlewm/20190621/content-1064445.html

Careful use of machine learning helps to promote the relationship between patients and doctors.

In 2013, IBM offered the University of Texas MD Anderson Cancer Center a bold idea to build an artificial intelligence-based platform, IBM Watson, that could use digitization to fight cancer, one of mankind's most abominable diseases.

But in less than four years, the groundbreaking project fell apart. Although the development of Waston has been full of ups and downs, it also shows us how machine learning technology can one day complement or even completely change what doctors do.

The question is not whether AI will enter the field of medicine, but how it will enter. At best, machine learning can use the collective experience of almost all clinicians to provide a doctor with the experience of millions of similar cases, allowing doctors to make informed decisions. In the worst cases, artificial intelligence can promote unsafe practices, expand social prejudices and show overcommitment, thus losing trust between doctors and patients.

Recently, Dr. Alvin Rajkomar and Dr. Jeffrey Dean of Google and Dr. Isaac Kohane of Harvard Medical School outlined the prospects and pitfalls of machine learning in medical practice in the New England Journal of Medicine (New England Journal of Medicine).

They believe that AI is not just a new tool, it is the basic technology to expand human cognitive ability, and has the potential to revolutionize health care for the better. Careful use of machine learning helps to promote the relationship between patients and doctors.

1. Diagnosis and treatment

In artificial intelligence and medicine, there is a lot of pressure on diagnosis.

Diagnostic tools based on artificial intelligence often detect potentially fatal lesions and diagnose skin cancer and retinal diseases through mammograms (now approved by FDA), even in their infancy, which some radiologists and pathologists may not see. Some AI models can even analyze mental symptoms or provide advice for referrals.

The latest advances in machine vision (machine vision) and transfer learning (transfer learning) have improved the diagnostic ability of computers. Although AI radiologists usually need a large number of annotated data sets to "learn", transfer learning allows previously trained AI to quickly acquire another similar skill. For example, an algorithm for training tens of millions of everyday objects in the standard repository ImageNet could be retrained for 100000 retinal images (a relatively small number of machine learning) to diagnose two common causes of vision loss.

More importantly, machine learning is well suited for analyzing data collected during daily care to identify possible future health conditions-minority health reports. These systems can help implement preventive measures, nip health problems in the bud and reduce health care costs. When sufficient quantity and quality of patient health longitudinal data are given, the prediction model established by AI is often more accurate than the original data obtained by medical practitioners through medical imaging.

The report's authors say doctors must be trained to collect the necessary information and input it into the AI prediction engine. We need to carefully analyze these models to ensure that they do not have charging incentives or situations in which advice cannot be given when there are no significant symptoms.

However, the AI model based on treatment data may only reflect the doctor's prescription habits rather than the ideal practice. A more helpful system must learn from carefully selected data to assess the impact of a treatment on a particular population.

It is difficult to achieve this. Some recent attempts have found it challenging to get expert data, update AI, or customize them based on practice. The authors conclude that giving medical advice with AI treatment is still at the forefront of the future.

2. Health care reform

The diagnosis is just the tip of the iceberg.

Perhaps more obvious is the impact of artificial intelligence on simplifying the workflow of doctors. In general, AI, such as intelligent search engines, can help select the necessary patient data. Other technologies, such as predictive types or voice dictation that doctors already use in daily practice, can simplify the tedious process of obtaining medical data.

The author emphasizes that we should not underestimate this special impact. Doctors are inundated with paperwork and take up valuable time with patients. Turning the main work into AI technology can improve efficiency and workflow, thus reducing burnout rate. More importantly, the data can in turn be fed back to the training machine learning model to further optimize patient care and carry out a virtuous circle.

Artificial intelligence is also the key to extending health care beyond clinics. For example, future apps will allow patients to take pictures of a rash and get an online diagnosis without having to run over for emergency treatment. Automatic triage can effectively transfer patients to appropriate nursing physicians. And after "seeing" billions of patients, machine learning can empower doctors to make better decisions, which may be the best hope for AI supplementary health care.

However, there is no data to support it in this particular case. The key now is to develop formal methods to test these ideas without harming doctors or patients.

3. Challenges and traps

Just as the development of IBM's Watson stumbles, the use of artificial intelligence in the medical community faces multiple challenges. The Theranos fiasco further paints a painfully clear picture of Silicon Valley's dogma "move fast and break things" being reckless and extremely dangerous in dealing with patients' health problems.

Medicine can often show the limits of machine learning. For example, in the absence of representative, diverse disease data sets, the AI model may be wrong or biased. This is part of the reason why IBM Watson crashed: it is very difficult to use a large enough annotated dataset to discover medical discoveries that are not yet known, even if possible.

However, the author believes that this is not a permanent obstacle. As long as the amount of data is large enough, the AI model will increasingly be able to deal with unreliable or variable data sets. Although not perfect, these models can be further refined by a smaller set of annotations, allowing researchers and clinicians to identify potential problems with the model.

Google Brain's work, for example, is exploring new ways to open the AI "black box", forcing algorithms to interpret their decisions. Interpretability is becoming more and more important in the clinical environment. Fortunately, famous AI diagnostics who publish articles in top journals today often have an inherent interpretation mechanism. While human experts can oversee the development of artificial intelligence alternatives to reduce misdiagnosis, all parties should make it clear that zero medical error rates are unrealistic for both humans and machines.

Clinicians and patients using these systems need to understand the limitations of their optimal use, the authors say. Neither side should rely too much on machine diagnosis, even if it becomes habitual and mundane.

Currently, we are limited to models based on historical data sets; the key in the next few years is to build forward-looking models that clinicians can evaluate in the real world, while dealing with the complex legal, privacy, ethical and regulatory issues faced by obtaining and managing large AI data sets.

The author is "cautiously optimistic" about this and expects some carefully scrutinized early models and cultural changes driven by economic incentives and value-based health care ideas in the coming years.

Finally, machine learning doesn't take anything away from doctors. On the contrary, the warmth of the doctor, her sensibility, sensitivity and appreciation of life will never go away. It will only be replenished.

"it's not about machines and humans, it's about optimizing human doctor and patient care by taking advantage of AI," Kohane said.

Reference: https://singularityhub.com/2019/04/25/how-ai-can-tap-into-the-collective-mind-to-transform-healthcare/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.