A recent study published in the European Journal of Cancer raises concerns about the lack of racial and ethnic diversity in datasets used for AI-driven mammogram interpretation. This underrepresentation could compromise the fairness and equity of AI models.

Key Findings:

• The study indicates that while AI has potential to improve mammogram interpretation, the diversity of datasets and the backgrounds of researchers involved in AI model development are critical.
• Researchers conducted a scientometric review of 5,774 studies published between 2017 and 2023, focusing on those that utilized mammograms for breast cancer detection.
• Out of these, only 264 studies met the inclusion criteria, with a notable increase in studies from 28 in 2017 to 115 in 2022-2023, marking a 311% rise.
• However, only 0-25% of these studies reported race or ethnicity, predominantly identifying patients as Caucasian.
• Most patient cohorts were from high-income countries, with no representation from low-income settings, and a gender imbalance was noted among authors.

The authors concluded that the lack of diversity in both datasets and researcher representation could undermine the generalizability and fairness of AI-based mammogram interpretation.

Implications for Breast Cancer Care:

Algorithms primarily trained on Caucasian populations may lead to inaccurate outcomes for underrepresented groups, potentially worsening existing disparities in patient outcomes.

The authors emphasized the importance of:

1. Prioritizing diversity in dataset collection.
2. Fostering international collaborations that include researchers from lower and middle-income countries.
3. Incorporating diverse populations in clinical research.

Related Developments in AI and Cancer Care:

In February, Google partnered with the Institute of Women’s Cancers to explore how AI can enhance cancer care and support researchers. Their focus includes:

• Forecasting cancer progression and relapse likelihood.
• Developing more accurate treatments for hard-to-treat women’s cancers, such as triple-negative breast cancer.

In 2024, Owkin collaborated with AstraZeneca to create an AI tool for pre-screening gBRCA mutations in breast cancer from digitized pathology slides, aiming to improve access to testing.

Additionally, Lunit and Volpara Health have joined forces to enhance early cancer detection and risk prediction through AI-powered solutions.

In May, Lunit acquired Volpara, integrating its breast health platforms into its AI tools for breast cancer detection. Lunit also signed a three-year agreement with Capio S:t Göran Hospital to provide AI-powered mammography analysis software, enabling the analysis of breast images for approximately 78,000 patients annually.