今日推荐英文原文：《Machine Learning: Diagnosing bots save lives》
今日推荐英文原文：《Machine Learning: Diagnosing bots save lives》作者：Hely Herranen
Machine Learning: Diagnosing bots save livesWhile the wait for the doctor’s office lengthen, medical data doubling every third year and healthcare costs growing, there is a shortage of medical professionals. Machine Learning provides an answer to many challenges in present-day medicine. Machine Learning Systems are fast and accurate at diagnosing.
This article discusses the problems Machine Learning can solve in the medical field. The requirements for Machine Learning Systems are addressed. The requirements are gathered from the referred article, Machine Learning for Medical Diagnosis: History, State of the Art and Perspective (Artificial Intelligence in Medicine), written by Igor Kononenko. In addition, this report introduces the supercomputer, IBM Watson, it’s history and arrival in the healthcare. In the last part, the report presents the diagnosing chatbots, Machine Learning Applications that focus on speed, accessibility, and accuracy. The goal of this article is to elaborate how diagnosing bots can save human lives. A couple of examples, Babylon Health and Woebot and studies on them, are introduced. The studies include: Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial carried out by the University of Stanford, and another study: A comparative study of artificial intelligence and human doctors for the purpose of triage and diagnosis written by Fitzpatrick KK, Darcy A, Vierhile M.
Machine LearningThe term Artificial Intelligence (AI) refers to intelligent computers, and a basic requirement to be intelligent is the ability to learn. Consequently, Machine Learning, a subfield of AI, is a very significant element in the project of making computers intelligent. Machine Learning uses statistical techniques to give computer systems the ability to learn from data, without being explicitly programmed. It evolved from pattern recognition and computational learning theory. Pattern recognition is automated recognition of patterns and regularities in data. Computational learning theory is the design and analysis of machine learning algorithms. In this part, the challenges Machine Learning can facilitate and the requirements for Machine Learning Systems are explained.
Why we need diagnosing systems?The world will be short of 12.9 million health-care workers by 2035. The shortage of doctors and other medical professionals is a threat, computer systems can assist in by speeding up the diagnosis and treatment processes. The patients would be able to describe their symptoms to a bot before seeing the doctor, who would then have all the information already needed to make good decisions.
According to IBM the amount of medical data doubles every third year. It makes the physicians unable to keep up with the latest research. Computers can manage massive amounts of data to help with the task of maintaining the latest information. Also, there is a challenge to expertise in medical fields. Patients are treated by general practitioners to all kinds of diseases. A Maschine Learning System can provide the expertise for the practitioners.
One challenge that matters is the speed of diagnosis and treatment. When deadly tumors are growing rapidly, it is a matter of life and death, or when a suicidal teenager goes into a waiting line to see a psychiatrist, the time is of the essence.
The patients will be able to get a ”second opinion” from the system, either encouragement that the doctor’s diagnosis is correct, or it may find the rare diseases, the physician was not able to diagnose.
Requirements for Machine Learning systems in healthcareTo be effective and reliable in medical diagnosis, there are requirements a Machine Learning System must meet. The following requirements include good performance, the ability to appropriately deal with missing data and with error in data, the transparency of data  and the recovery from biased data.
Good performance is the accuracy in which the system is able to produce the correct answers. The accuracy must be as high as possible. In some cases, the system performs better than human doctors do.
Dealing with missing data refers to the fact that for example, some relevant patient information may be missing. The algorithms must be able to appropriately deal with such incomplete descriptions of patients. Also, errors in the data can produce a problem. All data includes errors and uncertainty.
The conclusions the system comes to should be transparent, in a way, that it can be shown, how it ended up with the answers. All the decisions should be explained and backed up with research and data that was used. The physicians have to able to analyze and understand the knowledge generated by the system.
Another requirement is the recovery from biased data. Since humans train Machine Learning systems, the systems sometimes inherit unfortunate features like racism. The system is only as good as the data, it is trained on, the quality matters.
IBM Watson HealthcareIBM Watson Healthcare has developed into an intelligent system that helps doctors make informed decisions in cancer treatment. From winning a television quiz show to assisting doctors in diagnosing, the supercomputer developed by IBM has come a long way. In this part, we will discuss the development of the supercomputer and how it is used in medical diagnosis.
Supercomputer history and developmentTo many, IBM’s Watson is famous for winning the 2011 Jeopardy! quiz show against the best players in the history of the show. Succeeding in Jeopardy! requires an understanding of language, even humor, which is difficult for computers. However, Natural Language Processing (NLP), stored information and statistical analysis helped Watson to create the winning answers. Watson is not simply smart but growing more intelligent continuously by learning from success, failure and user feedback.
During the time of the Jeopardy! competition, Watson needed a separate room. Three years later, Watson had shrunk to the size of three pizza boxes and increased its processing speed by 240%. This can be explained by Moore’s Law, which means that the number of transistors placed in an integrated circuit doubles approximately every two years. Ergo, computers decrease in size and become more effective exponentially. There has been some speculations whether there will be an end to Moore’s Law. Still, we are moving towards an artificially intelligent future in a rapid speed.
Watson in HealthcareHealthcare professionals need to use vast amounts of data, such as patient history and the latest research, to diagnose their patients. Watson assists in the task by analyzing all the possible data. When a new patient comes to an assessment to the clinic, the doctor inputs all the critical medical information into Watson. Next, the supercomputer analyzes all the data, collects a list of hypothesis and recommends a treatment based on the patients’ medical information, best practices guidelines, latest worldwide research, and historical cases. All the data that Watson uses can be seen in detail, for example citations from researches. That is an endless source of information from where to conclude, much more than a human doctor can keep in working memory at once. The volume of medical research available doubles every three years which makes human doctors job to keep up with the latest very difficult. In addition, there are various side-effects of different drugs taken together, that is difficult to keep in working memory all of this information to come to a diagnosis or a treatment decision. Diagnosing is pattern recognition, which is what Machine Learning Systems do best and were made for.
In 2012, IBM partnered with Memorial Sloan Kettering Cancer Center to bring Watson into healthcare focusing on breast and lung cancer. The same year, the treatments Watson has suggested, went under a test by Texas MD Anderson Cancer Center, and they were compared to the suggested treatments of human physicians. The results overall accuracy were 82.6 %.
Applications for Machine Learning in diagnosingThere is an increased demand for services in healthcare. In addition, patients wish to gain control of their health decisions. A solution for these problems comes from the field of Machine Learning applications as diagnosing chatbots. Diagnosing bots are easily accessible to anyone and may provide a relief to a person in desperate need of advice. They reduce unnecessary visits to the hospital when the symptoms can be fixed at home, and they encourage when it is really the time to go to the hospital. The systems in practice are chatbot applications used via mobile or browser, easy to download or navigate to by anyone who has access to a computer. Entering the symptoms and answering the bots questions gives the patient a list of hypothesis and recommended action. The diagnosing bots are reliable, fast and accurate, and above all, they have the ability to save lives with their advice and encouragement. Next, a couple of examples and studies are introduced.
Babylon HealthBabylon’s mission is to put an accessible and affordable health service in the hands of every person on earth.
A study was made comparing Babylon Health to human doctors and the results suggest that the accuracy is comparable. The results also suggest that the treatment recommendations made by Babylon Health were in comparison even safer than human doctors recommendations. This proves that even not so high-end solutions like chatbots, can do the same tasks as doctors can.
WoebotWoebot is a free, therapy app, that focuses on mental issues. Woebot prompts the patient every day asking questions about their mood and how they are doing. A study was made to research if the application has any effect on depression.
In an unblinded trial, 70 individuals suffering from depression symptoms were recruited online from a university community social media site and put into two different groups. The other group had access to Woebot and the other one did not. The results of the study proved that the group that had access to Woebot reduced their depression symptoms significantly. A depressive person with suicidal thoughts can benefit enormously from talking to a chatbot. The bot knows to ask the right questions and gives good recommendations on what the person could do to improve their mood.
Use of AI necessary to surviveThe use of Artificial Intelligence in medical diagnosis is not only growing but necessary due to medical data doubling every third year, doctor shortage, healthcare costs rising and when the time is critical for treatment. Diagnosing is pattern matching which is what Machine Learning algorithms can do very precisely and effectively, better and faster than human doctors can. The systems are reliable and accessible. They increase the speed for treatment and can make a huge difference in helping patients find the right solutions for them.
There is no doubt, that the systems can do much good and while operating at the level of human doctors, it is enough to start their utilization. In the future, there will be many more tools to help medical professionals and patients to understand their symptoms better and conclude to the right diagnosis. The right diagnosis is crucial for saving a persons life.
- Igor Kononenko, Machine Learning for Medical Diagnosis: History, State of the Art and Perspective (Artificial Intelligence in Medicine, 2001), p. 89–109.
Susan Doyle-Lindrud, Watson Will See You Now: A Supercomputer to Help Clinicians Make Informed Treatment Decisions (Clinical Journal of Oncology Nursing, 2015), p.31–32
Qeethara Al-Shayea, Ghaleb El-Refae and Saad Yaseen, Artificial Neural Networks in Medical Diagnosis (International Journal of Behavioural and Healthcare Research, 2013), p. 45–63
Brijesh Verma and John Zakos, A computer-aided diagnosis system for digital mammograms based on fuzzy-neural and feature extraction techniques (IEEE Transactions on Information Technology in Biomedicine, 2001), p. 46–54
R. R. Schaller, “Moore’s law: past, present and future,” in IEEE Spectrum, vol. 34, no. 6, pp. 52–59, 1997
Understanding Machine Learning: From Theory to Algorithms, Shai Shalev-Shwartz and Shai Ben-David, Cambridge University Press, 2014
Digital Healthcare EcoSystem, Uday Kiran Kotla & Ginni Jain, Whitepaper by Infosys, 2018
A Universal Truth: No Health Without a Workforce, Third Global Forum on Human Resources for Health Report, Global Health Workforce Alliance and World Health Organization, 2013
Fitzpatrick KK, Darcy A, Vierhile M, Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial, JMIR Ment Health 2017;4(2):e19
A comparative study of artificial intelligence and human doctors for the purpose of triage and diagnosis, Razzaki et al. Babylon Health, School of Public Health, Faculty of Medicine, Imperial College London, Northeast Medical Group, Yale New Haven Health, Division of Primary Care and Population Health, School of Medicine, Stanford University