News and Events

Stroke Diagnosis and Prognosis with AI

Publication Date: 4/14/2024 9:00 AM

Physician-scientists from Rambam Health Care Campus (Rambam) in Haifa, Israel, are using clinical data and AI to diagnose and treat stroke patients. After a year-long study, in collaboration with scientists from the Technion-Israel Institute of Technology (Technion), their research is proving the ability of an OpenAI platform to diagnose strokes and deliver a 90-day prognosis for these patients.

Dr. Shahar Shelly. Photography: Rambam HCC.
Dr. Shahar Shelly. Photography: Rambam HCC.

In every discipline, OpenAI is being used to gather data, guide, and solve problems. In healthcare, it is proving to be lifesaving.

For the last year, Dr. Shahar Shelly, chair of Neurology at Rambam, and his colleagues have been delving into acute ischemic stroke management. Together with Technion scientists, they set out to confirm whether information obtained from GPT-4, a powerful AI platform, can enhance clinical decision-making and predict mortality rates in stroke patients.

In a retrospective study, the two institutions used existing data originally recorded for reasons other than research. They analyzed 100 consecutive cases of patients with acute stroke symptoms who had been admitted to Rambam’s Green-Wagner Department of Emergency Medicine.

Dr. Shelly explains, “We explored GPT-4’s capability in providing treatment recommendations and predicting mortality and compared its performance with expert opinions and real-world treatment outcomes.”

Existing machine learning models rely heavily on structured-data—rows and columns of clinical information—to process the data and gain insight. This study reveals GPT-4’s ability to process unstructured-data, such as images from PET, CAT, and MRI scans, X-rays, ultrasounds, physician’s notes, and more.

Dr. Shelly asserts that GPT-4 is a notable advancement, and says, “Results indicate that GPT-4’s ability to predict 90-day mortality rates surpass contemporary machine-learning models trained for this purpose.” In conclusion, he affirms, “GPT-4 aligns closely with clinical standards and has significant potential as a clinical tool. Our research will impact clinical practice and the application of AI in healthcare.”

The research manuscript, “Evaluating GPT-4 for Clinical Decision Support in Ischemic Stroke Management,” has been submitted to a leading, peer-reviewed medical journal for publication.

This collaborative study is one of many groundbreaking investigations that Rambam and the Technion conduct. Rambam’s Department of Neurology applies proven research, the newest and most advanced neurology technologies, drug treatments, and procedures to improve patient outcomes and save lives.