In January of this year (2017), the FDA granted the first approval for commercial use of a cloud-based artificial intelligence (AI) program in healthcare. The approval was granted to Arterys, a machine learning platform that was designed for use in cardiology. The program works by analyzing MRI imaging of the heart to accurately measure the volume of each ventricle, allowing physicians to precisely assess heart function. In order to receive this approval, Arterys had to demonstrate to the FDA that the results of its AI algorithm are at least as accurate as those obtained by a human clinician.¹

Arterys can conduct this assessment in 15 seconds on average, whereas a physician would be expected to take 30 minutes to an hour to perform the same task.¹ The program was able to achieve this speed while maintaining human level accuracy after being given a training set of only 1,000 patient cases to learn from. The cloud-based system will automatically receive data from all future patient assessments which will help it to further improve its accuracy.

What is AI?

Artificial intelligence comes from a branch of computer science known as machine learning. It’s a field where computers are designed to teach themselves how to perform tasks, given only basic rules that define the successful completion of the task. This is very different from most programs that run on our computers or smartphones which work by following strict, predefined, instructions that are executed under the direction of a human operator (the user). AI, on the other hand, is designed to perform tasks autonomously, taking out the human operator altogether.

How it works

AI works by utilizing neural networks, an information processing architecture that was modeled after how neurons in the human brain communicate. Using neural networks, an AI system can analyze sets of training data, which are provided by developers, in order to teach itself how to distinguish a correct result from an incorrect one. With sufficient training, an AI can utilize its newly learned insights to analyze sets of data that it has never seen before. The AI continues to improve its accuracy as it analyzes more new data. This process is called reinforcement learning.

AI’s role in our daily lives

Whether you realize it or not, you likely interact with AI algorithms every day. AI powers the search engines you use, populates your social media feeds and curates the recommendations you see when shopping online or browsing Netflix, YouTube or Spotify.² It’s also the brains behind virtual assistants such as Siri, Google Now and Amazon’s Alexa.

AI’s place in healthcare

AI has several characteristics which make it particularly well suited for applications within the healthcare field. Here are a few examples that illustrate why this is the case:

  1. AI excels in performing routine, predictable tasks: This makes AI ideal for use in diagnostic applications which involve standardized procedures and require a consistent assignment of diagnoses and grading of disease severity.
  2. AI development requires large amounts of quantitative data: Many clinical tests and procedures yield or rely upon quantitative data which has led to the aggregation of large data sets in health care. This data can be used to train AI systems to perform various tasks.
  3. AI excels in tasks where discriminating fine detail is the goal: The ability to discriminate details requires training and experience. A human physician can take a decade or more to complete their training and begin gaining experience. An AI system, on the other hand, can utilize the cloud to instantly share it’s newly learned insights with other AI’s that are performing the same task. This enables AI programs to gain many lifetime’s worths of clinical experience in a relatively short time.

Advantages of AI over human clinicians

Speed: The speed of an AI program is only limited by the computing power that it has access to. Because of cloud computing, internet connected AI programs will have access to large data centers with immense computing capacity. In contrast, even the most efficient doctor is limited by the number of neurons that can fit inside the human skull. This means that AI will be able to analyze a patient case in a fraction of the time that is needed by a human clinician.

Productivity: An AI system can analyze data 24/7. Computers don’t sleep and they don’t need time off for lunch breaks, weekends or vacations.

Cost: An AI algorithm can be cloned indefinitely and can be used to treat millions of patients simultaneously which allows for the spreading of R&D costs among many patients. This makes AI much more cost effective at performing a given task compared to a team of human doctors who each require compensation via salaries, paid time off, health insurance, etc.  

Accuracy and precision: AI programs don’t experience fatigue and they don’t get distracted which makes them immune to medical errors resulting from these factors which human clinicians are susceptible to.

Possibility of new breakthroughs: AI has the potential to unlock new insights relating to the diagnosis and management of disease because it approaches the problem from a different perspective. AI programs don’t follow current protocols or methods of analysis used by human clinicians. Rather, they derive their own knowledge by cross-referencing large sets of data and extracting patterns from it in ways that humans cannot.

The big picture

AI is set to play an increasingly large role in our daily lives in the coming years and it will undoubtedly disrupt many industries in the process. Integration of AI within the health care system is still in its infancy, but it’s clear that it has great potential to increase the clinician’s productivity, improve overall efficiency and contribute to better patient outcomes.

 

References:

  1. Marr, Bernard (2017, January 20). First FDA Approval For Clinical Cloud-Based Deep Learning. In Healthcare Retrieved from https://www.forbes.com/sites/bernardmarr/2017/01/20/first-fda-approval-for-clinical-cloud-based-deep-learning-in-healthcare/#6a1d6c4f161c

2.  Metz, Cade (2016, February 4). AI Is Transforming Google Search. The Rest of the Web Is Next. Retrieved from https://www.wired.com/2016/02/ai-is-changing-the-technology-behind-google-searches/