UVA has developed an AI to detect early sepsis. 2 Undergraduates paved the way

Applications of AI

the team behind the tools

Moore has spent much of his career looking for a way to fight sepsisRecognizing the diagnostic potential of AI, he sought to work with UVA experts.

He called his friend, Rich Nguyen of UVA Engineering.

Nguyen, an Assistant Professor in the Computer Science Department and also appointed to the Data Science Department, specializes in AI. He put together an interdisciplinary team.

“We aim for this collaboration to bring computer scientists and data scientists into clinical practice,” said Nguyen.

Two fourth-year students served as research assistants.

A statistics major, Edwards minors in computer science and social entrepreneurship. A Rodman Scholar, Boner gained experience as a software research intern and outside of Cisco before embarking on the sepsis project.

As part of their work, the students spent time in the medical ICU, circling the medical team under the direction of Dr. Tyson Bell and Kyle Enfield.

Behind the computer, “the team developed a data engineering pipeline,” Nguyen said. “They perform statistical and computational analyzes on large clinical data, allowing rapid experimentation with different machine learning models.”

The team also includes Joy Qiu, a 2020 Data Science graduate who works in the Center for Advanced Medical Analytics at UVA School of Medicine.

Computer Science graduate Matthew Pillari, a 2022 graduate, and Navid Jahromi, a 2021 graduate, have previously worked on this project. Pillari is currently a machine learning engineer at Imagen and Jahromi is a software engineer at Palantir Technologies.

What AI is learning

It is important to note that no medical decisions have yet been made based on this tool.

Because the AI ​​is still learning. And to learn, the AI ​​is immersed in a vast archive of biometrics. The data is basically replayed as if it were real time from the beginning of the patient’s admission.

“We feed AI with massive datasets,” says Boner. “The model is learning to match these data to determine if a patient has or does not have a bloodstream infection. So the AI ​​is learning patterns in the time series that we have and how the patient’s condition changes over time, which could be indicative of a bloodstream infection.”

The effort looks closely at specific types of patients, such as transplant recipients, Moore said. This is because their physiological responses to infection may differ.

As a result, there were some new discoveries.

“Transplant patients are immunocompromised,” the doctor explained. “This is because they are taking anti-rejection drugs.

“Our data suggest that they do indeed have a robust response. It may help us better identify bloodstream infections in the population.”

One of the dilemmas for physicians treating transplant patients is intervention and risk. For example, overuse of antibiotics can lead to antibiotic resistance and other unintended effects.

With AI able to read the nuances between individuals, more informed, more personalized care is possible.

Students learn while contributing

Like the technology itself, students have done a lot of deep learning.

Edwards said he learned about the challenges associated with using AI in medicine. She said being able to get first-hand insights from doctors and other medical professionals has helped her understand herself better. Instead, she wants it transformed into a tool.

“Our research is specifically focused on explainable artificial intelligence,” she said. “‘Explainability’ refers to the ability of an AI model to explain its behavior in human terms. Many of the most powerful machine learning models are so complex that they have a clear understanding of how to make No. Explainability is important for building trust in machine learning models, especially in clinical settings where lives are at stake.”

She added that she hopes to “continue working at the intersection of technology and social impact” wherever her career ends.

In addition to contributing to deep learning layers of AI, Boner writes: meeting materials With Nguyen and Moore as part of an undergraduate consortium.

“This project taught me, first and foremost, how to do research,” says Boner. “I have worked with both technical and non-technical researchers towards a common goal, and this has been very valuable.”

He is pursuing a PhD in Computer Science from Duke University, with a focus on interpretable AI for healthcare applications.

Moore praised both students for their many contributions to the project.

“Luisa and Zach are integral members of our research team,” Moore said. They are also very lively and bring fresh eyes and ideas to the problem of infection detection in the ICU.It was a pleasure to work with them and I learned a lot from them.”

test the AI

AI can now draw from the combined wisdom of 40,500 anonymized patient records, comprising 4.1 million laboratory readings.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *