Artificial Intelligence (AI) is everywhere you look — administration, smartphones, web searches, translating software, writing. It’s also at intersection of healthcare and higher ed, the very place the MGH Institute educates. 

Shuhan He is the program director for the MGH Institute of Health Professions’ Healthcare Data Analytics program, as well as an emergency medicine doctor and researcher in the field of AI. In this month’s IHP Interview with OSC’s Sean Hennessey, He talks about AI’s role in healthcare, higher education, what IHP students are learning about AI, and its limitless possibilities — and limitations. 

Talk about AI in healthcare. Where did AI begin, and where is it today? 

AI in healthcare has evolved significantly over the past two decades. Early applications included rule-based systems, reinforcement learning, and the first iterations of language models, but today, AI has become an integral part of many aspects of medicine. However, healthcare remains a highly regulated field where AI adoption requires extensive validation and oversight.

One of the most impactful areas of AI in healthcare has been radiology. AI-powered detection algorithms are now used to assist in identifying abnormalities in medical imaging. For example, when a patient undergoes a CT scan, AI can automatically flag potential concerns and triage those results, ensuring that critical cases are reviewed by radiologists more quickly.

It’s important to emphasize that AI is not currently used for independent diagnostics. Healthcare regulations demand a high level of accuracy, and AI-generated findings must always be reviewed and confirmed by human experts. In pathology, for instance, AI is often used to assist with preliminary screenings, helping pathologists detect abnormalities more efficiently before they verify and finalize diagnoses.

While AI is making meaningful contributions to healthcare, it is still in the early stages of direct clinical integration. However, ongoing advancements suggest that AI will continue to play an expanding role in improving efficiency and patient outcomes in the years to come.

Talk about the relationship or interaction between AI and data analytics.

Some people view AI and data analytics as separate fields, but in reality, AI models are just more complex extensions of traditional statistical models. When people think of data analytics, they often think of regression models or T-tests, but machine learning models are fundamentally data models as well — just more advanced and capable of handling larger and more complex datasets.

At its core, AI is a way to process and analyze data. Just like querying a database, you input a question, and AI processes it through intricate algorithms to generate an answer. These algorithms are all, in essence, data analytics models.

One of the key advancements AI brings to data analytics is its ability to unlock insights from unstructured data — particularly language. Traditionally, data analysis required programming skills to extract and interpret meaning, but AI-driven natural language models now allow us to interact with data more intuitively. This means we can ask more complex questions and derive insights in ways that were previously inaccessible.

Is it fair to say the MGH Institute’s Healthcare Data Analytics program is embracing AI as it instructs students? 

Absolutely. I often say that in a medical chart, there are two languages: English and numbers. We train healthcare professionals to interpret the English language when describing patient symptoms — such as noting that a patient appears diaphoretic and is clutching their chest. Then, they perform a physical exam and document numerical values — like lab results or vital signs.

Just as healthcare providers must learn to interpret written descriptions, they also need to understand how to interpret numerical data. Our program at MGH Institute teaches students to engage with this "number language" in the same way they would with English — to extract insights, recognize patterns, and make better clinical decisions.

How is AI being taught in the Healthcare Data Analytics program? 

We have a course called “AI Machine Learning.” We teach all the fundamental basics of AI, and next month we’re launching on our edX curriculum platform “Introduction to AI Machine Learning and Healthcare.” It's a free course for anyone who wants to learn about AI and machine learning and healthcare. We look at the overview of AI and its impact to healthcare, legal and ethical considerations, especially in privacy, the basics of data analytics and healthcare, and why that matters for machine learning. We look at all the algorithms you need to learn about machine learning and then we look at case studies. We think it's a very comprehensive course. It's open to the world. We want people to learn about it.

Do you see AI becoming the most dominant factor when it comes to data analytics, or is it already there?

I think it’s already happening, and it's only going to accelerate. That said, traditional statistical methods like T-tests and regression models remain foundational. Understanding concepts like linear regression helps build the intuition needed to grasp more complex AI models.

Ultimately, AI and machine learning are just evolutions of traditional data analytics. They allow for deeper insights, automation, and the ability to analyze much larger and more complex datasets.

Besides heading up the IHP’s Healthcare Data Analytics program and working as an emergency medicine doctor at Massachusetts General Hospital, you’re also a researcher there. Tell us about that.

My research spans a broad range of topics, but it all revolves around quantitative, algorithmic questions. I work on everything from how people use emojis to novel ways of classifying patients at discharge.

One of my key focus areas is developing new biostatistical models — particularly those related to entropy, which measures diagnostic uncertainty. Our goal is to use AI to minimize entropy, helping to refine diagnoses by eliminating as much uncertainty as possible.

To put it simply, we use AI to optimize medical decision-making. Rather than using traditional written language, we train AI on diagnosis codes and structured medical data to improve predictive accuracy. It’s a similar approach to how machine learning models are trained in other industries but applied to clinical medicine.

What comes to mind when you look at all the possibilities of what AI represents and what AI is already doing? 

It'll change everything. It's hard to say exactly when, where, how, and why but the technology is incredible. When it comes to EHRs (electronic health records), one thing you'll notice is that the patient never speaks for themselves. The doctor describes the patient, then describes the exam. The lab work is there about the patient, and then the doctor explains the diagnosis. I think as you have more language models, you'll have more direct, patient data to directly enter the EHR so that you can do better direct diagnosis. 

Another major area of interest is computer vision and its role in clinical assessments. A physician can often glance at a patient and quickly assess whether they appear critically ill or stable — something we refer to as human phenotyping. This visual recognition process is a fascinating challenge for AI. If we can teach AI to interpret visual cues as accurately as an experienced physician, it could revolutionize how we diagnose and triage patients.

Of course, real-world clinical practice involves more than just vision — it includes touch, smell, and other sensory inputs that AI still struggles to replicate. But in areas dominated by language and structured data, we’re already seeing significant advancements.

AI and machine learning are rapidly transforming healthcare, and those who don’t engage with it now risk being left behind. This is the future — it's time to get on board.

Do you have a story the Office of Strategic Communications should know about? If so, let us know.