Sunday, April 14, 2024
HomeMedical TechnologyMedicine with an AI

Medicine with an AI

- Advertisement -

Google’s New AI Is Acquiring Patient Diagnostic Skills With AMIE, an AI model, the DeepMind team turns to “Medicine with an AI”.

Even under the best of circumstances, navigating the health care system as a patient can be difficult, whether you’re deciphering diagnosis full of technical terms or figuring out which doctors to visit next. In a similar vein, physicians frequently have demanding schedules that make it challenging to provide each patient with individualized care. These problems are especially made worse in places with a shortage of medical professionals and facilities.

Since IBM’s Watson made its debut more than ten years ago, researchers have been working toward the ideal of bringing AI inside the doctor’s office to help with these issues, but progress has been sluggish. These days, ChatGPT and other large language models (LLMs) may be able to revive such goals.

In a recent preprint document that was released on arXiv on January 11th, the Google DeepMind team developed a new AI model named AMIE (Articulate Medical Intelligence Explorer). During a wellness visit session, the model might gather data from patients and give concise explanations of medical concerns.

Lead author of the most recent article is Google AI researcher Vivek Natarajan. Although he acknowledges that AMIE isn’t meant to take the position of human doctors, he does think that a comparable AI may be useful in supporting both medical professionals and patients.

As part of or in addition to their treatment travels, “there may be scenarios where people might benefit from interacting with systems like AMIE,” adds Natarajan. “These include providing a valuable second opinion and simplifying explanations in local vernaculars in order to better understand symptoms and conditions.”

This might therefore open the door for medical AI to achieve superhuman diagnostic capabilities.

- Advertisement -

The AI Patient Actor software was developed by Thomas Thesen, an associate professor of medical education at Dartmouth’s Geisel School of Medicine, to assist in preparing medical students for a variety of patient scenarios. Though he believes AI will become more prevalent in healthcare, he does not think it will take the place of human physicians’ skills.

Thesen predicts that over the next ten years, artificial intelligence (AI) will help physicians more and more by expediting their tasks and assisting with certain, constrained diagnostic procedures. “Yet, the final diagnosis and treatment plans will depend on the expert judgment of a qualified physician.”

Natarajan and colleagues began feeding the AI real-world medical texts to bring AMIE up to speed without putting it through medical school. These texts included transcripts of nearly 100,000 real physician-patient conversations, 65 clinician-written summaries of medical notes from intensive care units, and thousands of medical reasoning questions from the US Medical Licensing Examination.

However, Natarajan notes that even with this data, AMIE would not have been successful on its own because the data are often noisy and only cover a tiny portion of possible medical circumstances. In order to close these gaps, the group also employed a simulated diagnostic environment, which provided AMIE with two distinct “self-play” cycles for learning from its errors.

According to Natarajan, “the environment included two self-play loops: a ‘outer’ self-play loop where the set of refined simulated dialogues were incorporated into subsequent fine-tuning iterations, and a ‘inner’ self-play loop where AMIE leveraged in-context critic feedback to refine its behavior on simulated conversations with an AI patient simulator.” “Thereby, establishing a positive continuous learning cycle, the resulting new version of AMIE could take part in the inner loop once more.”

Even while Natarajan emphasizes that actual human experience in medicine cannot be replaced, AMIE does have some advantages over human physicians thanks to this training paradigm. A human doctor, for instance, could only visit 10,000 patients throughout their career; but, in a matter of training cycles, AMIE might “see” that many patients.

According to Natarajan, “this in turn potentially provides a pathway for medical AI towards superhuman diagnostic performance.”

In a blind and randomized controlled experiment, Natarajan and colleagues compared AMIE to 20 human primary-care physicians to assess how well it performed in consultation with patient actors in Canada, India, and the UK. Live texting was used for 149 distinct consultations, which were assessed by real professionals as well as the patient actors.

“The final diagnosis and treatment plans will always depend on the expert judgment of a qualified physician.”

A number of criteria were used to evaluate the consultations, including management planning, openness and honesty, perceived empathy, and diagnostic accuracy. When compared to their human counterparts, AMIE offered “greater diagnostic accuracy and superior performance,” according to the evaluations of both the patient actors and the professionals. These outcomes aren’t always as clear-cut as they seem, though.

To begin with, these consultations were conducted using the same live text-based chats that are usually used for LLM communication. But this format differs greatly from the kind of in-person communication that doctors are accustomed to, which might be advantageous for AMIE. The researchers discovered that AMIE also had a tendency to write somewhat lengthier replies than human doctors, which they think patients would see as more time-consuming and, thus, considerate and sympathetic.

In the future, Natarajan and associates would like to add multimodal sources, including video conversations, to AMIE’s capabilities. In order to properly prepare AMIE for the real world, the team will also examine issues with equality, fairness, and adversarial testing.

Regarding the human doctors who are waiting for AMIE to arrive, Thesen thinks it’s critical that they become ready for how this technology may transform healthcare.

Thesen asserts that it is the duty of medical schools to include AI literacy into their curricula. “This includes being aware of the ethical ramifications so that future medical professionals can use AI sensibly and safeguard the welfare of their patients as it becomes more integrated into clinical practice.”

- Advertisement -
Next article


Please enter your comment!
Please enter your name here

Most Popular