Commentary

Video

Leveraging AI to Reduce Breast Cancer Risk

Mikael Eriksson, PhD, epidemiologist at Karolinska Institute in Sweden, discusses the development and implementation of an artificial intelligence (AI) model to reduce bias and improve breast cancer prevention.

Mikael Eriksson, PhD, epidemiologist at Karolinska Institute in Sweden, explains how to reduce bias within artificial intelligence (AI) models using large data sets that can help prevent breast cancer. He outlined the development process and clinical pathway for the AI model.

Eriksson presented a long-term image-derived AI risk model for primary prevention of breast cancer at the San Antonio Breast Cancer Symposium 2024 between December 10 to December 13.

This transcript has been lightly edited for clarity.

Transcript

What steps were taken to mitigate potential biases in the AI model, such as algorithmic bias or data bias?

The biggest problem with the AI models is that they can learn very well but they can also start to learn the specific type of image data that are using it. The images, for instance, that are particular to that brand, the AI can start to learn that pattern so we need to have a big mix of data from different vendors, different screening routines, different caregivers, different countries, and different ethnicities.

When we have this very rich data set, then we can lower the bias of the AI learning specific features for one specific setting and it really captures the causal relationship between the image features and the outcome of breast cancer.

How do you envision the integration of this AI model into routine clinical practice?

Yeah, so we have a typical phase. We start by developing the model and that is done in retrospective data and as we are epidemiologists, we are very keen on having large data sets and that coincides very well with AI because it really needs large data sets. Then we go into the prospective setting where we test the data for completely new scenarios that the AI has not been seen before.

Then we come into the clinical trials where we really, in the best possible way, can randomize women into different arms and then test the AI model in the best possible ways. After that comes the potential clinical approval of the model.

Related Videos
Toby Maher, MD, PhD, professor of clinical medicine, Keck School of Medicine at USC
Nini Wu, MD, Navista
Fred Locke, MD, Moffitt Cancer Center and Research Institute
Dr Emma Achola-Kothari
Matthew Viggiano, MD, internal medicine resident, Temple University Hospital
Krunal Patel, MD, pulmonary and critical care fellow, Temple University Hospital
M. Bradley Drummond, MD, MHS, professor of medicine, University of North Carolina at Chapel Hill
Dr Margrit Wiesendanger
Sanjay Ramakrishnan, MD, senior lecturer, University of Western Australia
Michael Goulet, DO, pulmonary and critical care fellow, Temple University Hospital
Related Content
AJMC Managed Markets Network Logo
CH LogoCenter for Biosimilars Logo