Researchers want to use an AI system to better predict the risk of transplant rejection. The system is intended to support human experts in the diagnosis.
Organ transplants can save lives if the body accepts the new organ. However, if the immune system classifies the tissue as foreign and fights it, this can lead to several symptoms over different periods of time that cannot always be clearly attributed to rejection. Chronic rejection can even last for years.
Because of the many possible reactions, experts arrive at different diagnoses when assessing the degree and severity of rejection. These, in turn, can lead to delays in treatment or inappropriate drug dosing. This is where medical AI should assist as a tool by promoting consensus.
AI diagnosis on a human scale
A team from Harvard Medical School and Brigham and Women's Hospital presents the AI system "Cardiac Rejection Assessment Neural Estimator (CRANE)" in the journal Nature Medicine. It is designed to help physicians diagnose rejection earlier and more accurately.
The researchers trained the AI system using Deep Learning to detect, subdivide, and classify transplant rejections. Like many medical diagnostic systems, Crane works visually by examining patterns on medical images. The research team used thousands of pathology images from more than 1,300 heart biopsies at Brigham and Women's Hospital as training material.
In a pilot study, the AI system demonstrated diagnostic capability on par with human experts, according to the research team. In cooperation with them, the system reportedly provided more matches and reduced assessment time.
"Our retrospective pilot study demonstrated that combining artificial intelligence and human intelligence can improve expert agreement and reduce the time needed to evaluate biopsies," said Faisal Mahmood, the study's lead author and an assistant professor of pathology at Brigham and Women's.
Clinical trial to follow
The system is designed to be used by multiple experts or as a complement when few pathology specialists are available, Mahmood said. The results so far will lay the groundwork for a clinical trial to determine the exact benefits of AI for successful heart transplants, Mahmood said.
“Throughout the history of medicine, diagnostic assessments have been largely subjective,” said Mahmood. “But because of the power and assistance of computational tools, that’s beginning to change."
The researchers intentionally tested the AI on samples from patients in the U.S., Turkey, and Switzerland to maximize differences between populations, sample preparations, and slide scanning instruments during validation. Many unknown variables are stress tests for the robustness of artificial intelligence.
In an interactive browser application, the team provides sample diagnoses from Crane.
Read more about AI:
- AI: Six characteristics for good algorithms in medicine
- Artificial Intelligence is the Future of Deterrence
- Artificial intelligence to fly US fighter jets in 2024