The current and future of machine learning in the medical field

In the big wave of artificial intelligence sweeping the world, machine learning technology applied in the medical field seems to have become the top topic of concern for the investment community and the press in the past three years.

Since the beginning of 2013, IBM's Watson has been used in the medical field; after defeating the world's best Go players, Google's DeepMind team decided to focus on the full use of this technology in the medical field.

Many of the most popular young start-ups in the machine learning (ML) industry have also invested a lot of energy in the medical field, including Nervanasys (recently acquired by Intel), Ayasdi (financing $9.4 million as of February 2016), and Sentient. Ai (financing $144 million as of February 2016) and Digital Reasing Systems (financing $36 million as of February 2016) are numerous.

Despite the excitement of the investment and research communities, we have found that most machine learning company executives have a hard time understanding what aspects of machine learning are now playing in the medical field. To this end, we are writing this article, not to outline the possible applications, but to focus on the current and future uses of machine learning in the medical field.

Machine learning is currently in the medical field

The following applications are not comprehensive, but show some of the impact of machine learning in the medical industry .

Medical image diagnosis

Computer vision has always been one of the most significant technological breakthroughs due to machine learning and deep learning, and is also a particularly active application area for machine learning in the medical community. Microsoft's InnerEye program (starting in 2010) is currently developing image diagnostic tools, and the team has released many videos explaining development trends, including machine learning for image analysis.

As deep learning becomes more prevalent and more data sources (including a wide variety of medical images) become part of the artificial intelligence diagnostic process, deep learning may play an increasingly important role in diagnostic applications.

However, deep learning applications have limited interpretation capabilities. In other words, a trained deep learning system cannot explain how it "predicts" predictions, even if the predictions are correct. This "black box problem" is more challenging in the medical world, and doctors don't want to make life-and-death decisions without a deep understanding of how the machine gives advice, even if it has proven to be correct in the past.

Treatment queries and suggestions

Diagnosis is a very complex process involving many factors: from the color of the patient's white to the food that is eaten in the morning – the machine is currently unable to correlate and interpret these factors. However, there is no doubt that simply acting as an extension of scientific knowledge can help doctors make the right considerations in diagnosis and treatment.

This is the goal that Memorial Sloan Kettering (MSK)'s oncology department recently worked with in IBM Watson. MSK has a wealth of data on cancer patients and the treatments that have been used for decades, drawing on the most effective treatments in the past to show and suggest treatment ideas or protocols to treat unique cancer cases in the future. Today, this smart enhancement tool is already in initial use.

Medical data collection

People are very concerned about bringing together data from a variety of mobile devices to aggregate and interpret more real-time health data. Apple's ResearchKit is designed to treat Parkinson's disease and Aspergol's syndrome by allowing users to access interactive applications (one of which uses machine learning to identify faces) for long-term assessment of physical condition. The application can feed daily progress data into an anonymous data pool for future research.

IBM is sparing no effort to access all the health data it can get, not only working with Medtronic to interpret diabetes and insulin data in real time, but also spending $2.6 billion to acquire medical analysis company Truven Health.

Although the Internet of Things provides a large amount of health care data, the industry seems to be trying to interpret this information and change the treatment in real time. Scientists and patients are optimistic that if the trend of bringing together consumer data continues, researchers will have more tools to overcome incurable diseases and unique cases.

Drug discovery

While the healthcare industry involves many stakeholders (hospital CIOs, doctors, nurses, patients, insurance companies, etc.), drug development brings relatively simple economic value to companies developing machine learning healthcare applications. This kind of application software also faces a group of relatively clear customers, who are usually financially strong, that is, pharmaceutical companies.

IBM's own medical applications have long been focused on drug discovery, and Google has joined the ranks of drug discovery, and a large number of companies are already financing and profiting through drug discovery through machine learning.