The vast majority of AI models used in medicine today are “limited specialists,” trained to do one or two tasks, such as scanning mammograms for signs of breast cancer or detecting lung disease on chest X-rays.
But the daily practice of medicine involves an endless variety of clinical scenarios, symptom presentations, possible diagnoses, and treatment puzzles. So if AI is to deliver on its promise to reshape clinical care, it must reflect that complexity of medicine, and do so with high fidelity, says Pranav Rajpurkar, an assistant professor of biomedical informatics at the Blavatnik Institute at HMS.
Enter general medical AI, a more evolved form of machine learning capable of performing complex tasks in a wide range of scenarios.
Similar to GPs, Rajpurkar explained, GP AI models can integrate multiple types of data, such as MRIs, X-rays, blood test results, medical texts and genomic tests, to perform a variety of tasks, from making complex diagnostic calls to supporting clinical decisions to choosing the optimal treatment. And they can be deployed in a variety of settings, from the exam room to the hospital ward, from the outpatient GI procedure room to the heart operating room.
While the first versions of general medical AI have begun to emerge, its true potential and the depth of its capabilities have yet to be realized.
“The rapidly evolving capabilities in the field of AI have completely redefined what we can do in the field of medical AI,” Rajpurkar writes in a recently published perspective in Naturein which he is co-senior author with Eric Topol of the Scripps Research Institute and colleagues from Stanford University, Yale University, and the University of Toronto.
General medical AI is poised to transform clinical medicine as we know it, but with this opportunity comes serious challenges, say the authors.
In the article, the authors discuss the defining characteristics of general medical AI, identify various clinical scenarios in which these models can be used, and chart the way forward for their design, development, and implementation.
Characteristics of general medical AI
The key characteristics that make generalist medical AI models superior to conventional models are their adaptability, their versatility, and their ability to apply existing knowledge to new contexts.
For example, a traditional AI model trained to detect brain tumors on a brain MRI will look at a lesion on an image to determine if it is a tumor. You cannot provide information beyond that. In contrast, a generalist model would look at a lesion and determine what type of lesion it is: a tumor, a cyst, an infection, or something else. He or she can recommend additional tests and, based on the diagnosis, suggest treatment options.
“Compared to current models, general medical AI will be able to perform more sophisticated reasoning and integrate multiple types of data, allowing it to create a more detailed picture of a patient’s case,” said study co-author Oishi Banerjee, a researcher associate in Rajpurkar’s lab, which is already working on the design of such models.
According to the authors, generalist models will be able to:
- Easily adapt to new tasks without the need for formal training. They will complete the task simply by asking them to explain it to them in plain English or another language.
- Analyze various types of data—images, medical text, lab results, genetic sequencing, patient records, or any combination of these—and generate a decision. By contrast, conventional AI models are limited to using predefined data types (text only, image only) and only in certain combinations.
- Apply medical knowledge to reason through never-before-seen tasks and use medically accurate language to explain your reasoning.
Clinical scenarios for the use of general medical AI
The researchers describe many areas where generalist medical AI models would offer comprehensive solutions.
Some of them are:
- Radiology reports.
The general practitioner AI would act as a versatile digital radiology assistant to reduce workload and minimize memory work.
These models could write radiology reports describing both abnormalities and relevant normal findings, also taking into account the patient’s history.
These models would also combine text narrative with visualization to highlight areas in an image described by text.
The models could also compare past and current findings on a patient’s image to illuminate telltale changes that suggest disease progression.
- Surgical assistance in real time.
If an operating team runs into a roadblock during a procedure, such as not finding an organ mass, the surgeon might ask the model to review the last 15 minutes of the procedure for flaws or oversights.
If a surgeon encounters an ultra-rare anatomical feature during surgery, the model could quickly access all published work on this procedure to provide real-time insights.
- Bedside decision support.
Generalist models would offer alerts and treatment recommendations for hospitalized patients by continuously monitoring their vital signs and other parameters, including patient records.
Models could anticipate impending emergencies before they happen. For example, a model can alert the clinical team when a patient is about to experience circulatory shock and immediately suggest steps to avoid it.
Ahead, promise and danger
Generalist medical AI models have the potential to transform healthcare, say the authors. They can alleviate physician burnout, reduce clinical errors, and speed up and improve clinical decision making.
However, these models come with unique challenges. Its strongest characteristics—extreme versatility and adaptability—also present the greatest risks, the researchers warn, because they will require the collection of vast and diverse data.
Some critical traps include:
- Need for extensive and continuous training.
To ensure models can change data modalities quickly and adapt in real time based on context and the type of question asked, they will need to undergo extensive training on diverse data from multiple sources and complementary modalities.
That training would have to take place periodically to keep up with new information.
For example, in the case of new variants of SARS-CoV-2, a model must be able to quickly retrieve key features on X-ray images of pneumonia caused by an older variant to contrast lung changes associated with a new variant. variant.
- Validation.
Generalist models will be exceptionally difficult to validate due to the versatility and complexity of the tasks they will be asked to perform.
This means that the model must be tested in a wide range of cases that it may encounter to ensure proper performance.
What this boils down to, Rajpurkar said, is defining the conditions under which the models work and the conditions under which they fail.
- Check.
Compared to conventional models, general medical AI will handle much more data, more varied types of data, and more complex data.
This will make it much more difficult for clinicians to determine how accurate a model’s decision is.
For example, a conventional model would look at an imaging study or full slide image when classifying a patient’s tumor. A single radiologist or pathologist could verify whether the model was correct.
By comparison, a generalist model might analyze pathology slides, CT scans, and medical literature, among many other variables, to classify and stage disease and make a treatment recommendation.
Such a complex decision would require verification by a multidisciplinary panel that includes radiologists, pathologists, and oncologists to assess the accuracy of the model.
The researchers note that designers could facilitate this verification process by incorporating explanations, such as clickable links to supporting passages in the literature, to allow clinicians to efficiently verify the model’s predictions.
Another important feature would be the construction of models that quantify their level of uncertainty.
- biases
It’s no secret that medical AI models can perpetuate biases, which they can acquire during training when exposed to limited data sets obtained from non-diverse populations.
Such risks will be magnified when designing general medical AI due to the unprecedented scale and complexity of the data sets required during its training.
To minimize this risk, generalist medical AI models should be thoroughly validated to ensure they do not underperform in particular populations, such as minority groups, the researchers recommend.
In addition, they will need to undergo ongoing audits and regulation after deployment.
“These are serious but not insurmountable obstacles,” Rajpurkar said. “Having a clear understanding of all the challenges early on will help ensure general medical AI delivers on its tremendous promise to change the practice of medicine for the better.”
—————————————————-
Source link