I am often asked, “What advances are we seeing when it comes to artificial intelligence (AI) in healthcare?”
This is a great question and one that I love discussing as a co-founder of InnoVet Health, an AI and digital services healthcare IT company. I was originally trained as a mathematician but got converted to health informatics and AI through a little project called ILIAD.
ILIAD was conceptualized in 1984 as I was starting my post-doctoral training. It was an innovative collaboration between the University of Utah and Intermountain Healthcare, a Utah-based, not-for-profit medical group of 33 hospitals. The entirety of the project lasted 10+ years, and represents the combined effort of a multi-disciplinary team under the leadership of Dr. Homer Warner, a cardiologist and renowned pioneer of medical informatics.
The vision was to develop a software system that helps diagnose difficult cases and, at the same time, teach medical students the art and science of medical decision-making.
It was an ambitious goal. A goal that took years of research and hard work leading to the development of ILIAD; software expert created to behave like a human medical expert.
A Brief Overview of ILIAD, a Medical Expert System
Essentially, ILIAD uses medical information, pre-engineered into a specific format, to estimate the likelihood of a diagnosis and recommend additional tests and, eventually, treatment options.
ILIAD is an artificial intelligence system based on a Bayesian algorithm; a statistical method that assigns conditional probabilities to events given other events have taken place and the likelihood of these events. ILIAD used big population health data analytics and medical research literature to estimate these input likelihoods. Bayes theorem calculated the output probability.
ILIAD associates each disease with a general prevalence probability and a list of diagnostic findings (symptoms, signs, lab tests, etc.) that are weighted in terms of sensitivity (likelihood in the disease population) and specificity (likelihood in the non-disease population). When a user enters a patient’s medical history in ILIAD, the system can generate a differential diagnosis as well as recommend additional tests to be done and treatment plan to consider.
In addition to helping diagnose difficult cases, ILIAD also teaches medical students how to make educated decisions throughout the patient workup process. Using its simulation mode and statistics knowledge base, ILIAD can generate synthetic patient cases and present them to medical students to identify the diagnosis, enabling students to ask questions, and propose diagnoses for ILIAD (the expert) to evaluate their relevance and score automatically.
This simulation mode with its built-in feedback loop provides medical students with case-based realistic learning and training opportunities, in an automated AI fashion.
ILIAD in Retrospect
The achievements made with ILIAD were enormous! The team was proud to see it used in over half of US medical schools and to be nominated as ‘best teacher of the year.’ It took an estimated 100,000 engineering hours to refine ILIAD’s performance to a satisfactory level. As a bonus, the ILIAD AI model demonstrated that human intelligence or expertise is not an intrinsic sophisticated algorithm but rather a direct result of acquiring extensive knowledge of a domain. Indeed, the team showed that, for instance, a cardiologist can be an expert in heart diseases but a novice in another domain, say infectious diseases. In other words, the cardiologist’s expertise is encapsulated in the amount of experience, data points, and knowledge in the heart domain, and not in a pure intelligent formula that can be applied to any domain.
Our conclusion is that AI in healthcare is doing well.
Stories like ILIAD illustrate the power of data and algorithms. Today’s medical data are vast and rich. Through open APIs, algorithms are easier to implement and connect to live clinical systems and Electronic Health Records (EHRs), thus, making these decision support systems more intelligent, more integrated, and more useful to providers and patients.
However, it’s important to keep in mind that ILIAD is just one approach to AI. By contrast we have seen other applications of artificial intelligence such as IBM Watson, which does not require pre-engineered knowledge with expert judgement and compiled statistics. Instead, Watson can consume raw information, like textbooks and journal articles to predict a specific result.
You can read more about other applications of AI in healthcare from our Senior Business Intelligence Lead, Ivan Castro, here.
We are excited to see AI in healthcare continue to develop, and as a company, we are proud to be part of these developments and offer this expertise to our clients!
You can read more about ILIAD and knowledge engineering in health informatics here!
By: Omar Bouhaddou, PhD (Co-Founder & CHIO of InnoVet Health)
InnoVet Health is an IT consultant company specializing in AI and business intelligence, digital services, and health interoperability founded by MIT-alumni & informatics experts. Learn more about us on our website or reach out on LinkedIn.