Dr. Arnab Roy writes on experimental AI, the development of which has come from a substantial progress in our understanding of what forms clinical expertise, and the translation of such expertise into cognitive models, leading it to become a vital part of front and back office, and clinical operation functions.

I like the definition that Wikipedia has - ‘Artificial intelligence (AI) in healthcare is the use of complex algorithms and software to estimate human cognition in the analysis of complicated medical data.’ At the same time, we should be aware of two different forms of AI that the experts are talking about – General AI and Narrow AI.

General AI is something like what the robots in science fiction movies represent – human-like and human independent. And for sure, we are far, far away from developing it. We are actually in the era of Narrow AI – a software model that works in a pre-determined, pre-defined and predictable manner often in a much more sophisticated way than a human can. Take, for example, the Amazon Alexa – it can calculate sums much faster than that is humanly possible, it answers questions appearing like a genius, it can order groceries for you, but it cannot fetch the groceries! Neither can it cook food for you or dust your house.

Similarly, the field of healthcare AI is seemingly wide—covering wellness to diagnostics to operational robots—but it is also narrow in that the AI can perform just a single task.

But is healthcare AI different from AI in other industries?

Yes, it is. Healthcare data is different from banking transaction data or weather data. In the same manner, AI in healthcare that uses healthcare data needs to be different from that in other industries. It has to be precise, personalised, accurate and above all - ethical.

The AI that is currently being used in the industry is 'narrow', designed to perform specific tasks in a far better and quicker way than humans and set to benefit the entire value chain of the industry, but the development process of which requires equal participation by doctors and IT professionals.

A bit on the history of AI

The term artificial intelligence was first coined by John McCarthy, an Assistant Professor of Mathematics at Dartmouth College in 1956 when he organised the first academic conference on this subject - Dartmouth Summer Research Project on Artificial Intelligence. Research in the 1960s and 1970s produced the first problem-solving program, or expert system, known as Dendral.

While it was designed for applications in organic chemistry, it provided the basis for the subsequent system MYCIN, considered to be the earliest artificial intelligence model in medicine. Many of the early efforts to apply artificial intelligence in medical reasoning were typically easy to create because they were in the form of "if/then" rules to reach a conclusion.

However, MYCIN and other systems such as INTERNIST-1 and CASNET did not achieve routine use by practitioners as the model outcomes were narrow and variable. Such rule-based models could not incorporate the myriad biological variations seen both in healthy and diseased states.

Cut to modern times

Given the difficulties encountered with rule-based systems, more recent efforts to use artificial intelligence in medicine have focused on programmes organised around models of disease, a bottom-up approach. Efforts to develop such programs have led to substantial progress in our understanding of what forms clinical expertise, the translation of such expertise into cognitive models, and the conversion of such models into experimental AI programmes. Kindly note, these programmes steadily improve through the identification and correction of flaws shown by exposing them to various clinical scenarios.

In the healthcare set up, AI is slowly becoming part of the front office, back office and the clinical operations functions.

Front Office

AI front desk assistant to record

• demographics
• clinical history
• family history
• sysmptoms
• appointments
• set up reminder service

Clinical Operations

• Robot assisted surgeries
• AI enabled interpretation of patient history, diagnostic test results and treatment regime
• Fetch clinical reference from cloud based Big Data or from back office hospital server
• Interpretation of x-rays, mammography, MRI, CT scan, biopsy images

Back Office

• dive into EMR and fetch similar case information for supporting clinical decision making
• developing prediction tools for chronic conditions and their complications e.g diabetes and retinopathy
• link up to information from patients' wearables and feed into personalised EMR
• billing and fraud detection

What lies ahead?

Artificial Intelligence holds great promise for healthcare improvement;  but developing AI models cannot lie solely with the technical experts from Computer Sciences domain– doctors’ involvement is a must in its development process. However, we must recognise that AI is not a flawless system as yet (and may never will be, like we humans) as many practitioners would expect it to be.

The AI models are the product of algorithms that have been or will be constructed on human cues and human generated data, and will, though inadvertently, inherit some of the biases of humans. So, however hard we try we may still not have an AI model which will be 100% agnostic process of objective data analysis and results. But the AI applications of the near future will for sure deliver great value to maintaining human health and should be prioritised and invested in, so that doctors focus more on the qualitative aspects of healthcare and upgrade themselves to the next level of clinical excellence.