Study: Racial, ethnic minorities are underrepresented in AI mammogram interpretation

According to a study published in the European Journal of Cancer, the fairness and equity of datasets for AI-driven mammogram interpretation might be jeopardized by the underrepresentation of racial and ethnic diversity.

While AI shows promise for enhancing how mammograms are interpreted, particularly in areas where resources are limited, the study’s authors found warning signs regarding the diversity of datasets and the representation of researchers in AI model development, which they said could  “affect the models’ generalizability, fairness and equity.”

For the study, researchers conducted a scientometric review of studies published in 2017, 2018, 2022 and 2023 utilizing screening or diagnostic mammograms for breast cancer detection to “train or validate AI algorithms.”

Of the 5,774 studies identified, 264 met the inclusion criteria. 

“The number of studies increased from 28 in 2017 to 2018 to 115 in 2022 to 2023–a 311% increase. Despite this growth, only 0–25 % of studies reported race/ethnicity, with most patients identified as Caucasian,” the study’s authors wrote. 

“Moreover, nearly all patient cohorts originated from high-income countries, with no studies from low-income settings. Author affiliations were predominantly from high-income regions and gender imbalance was observed among first and last authors.”

The authors concluded: “The lack of racial, ethnic and geographic diversity in both datasets and researcher representation could undermine the generalizability and fairness of AI-based mammogram interpretation.”

Furthermore, recognizing the disparities through diverse dataset collection and comprehensive international collaborations is crucial to guaranteeing fair advancements  in breast cancer care.

Study data revealed that algorithms focusing mostly on caucasian populations could result in inaccurate outcomes and the wrong diagnosis in underrepresented populations. Additionally, patient outcomes may be threatened and current disparities could worsen. 

“The fairness of these AI tools is called into question, as they risk systematically dis-advantaging certain racial, ethnic or socio-demographic groups. To mitigate these issues and ensure that the benefits of AI in BC imaging are equitably distributed, it is essential to prioritize diversity in dataset collection, foster international collaborations that include researchers from lower and middle-income countries and actively incorporate diverse populations in clinical research,” the study’s authors wrote. 

THE LARGER TREND

In February, Google partnered with the Institute of Women’s Cancers, founded by France’s cancer research and treatment center, Institut Curie, to study how AI tools can help address cancer, share science-based health information and support postdoctoral researchers with funding. 

The two entities looked into how AI-based tools can help forecast the progression of cancer and the likelihood of relapse for patients, with the goal of developing more accurate and successful treatments.

The researchers focused on hard to treat women’s cancers, including triple-negative breast cancer, an aggressive type of breast cancer that grows and spreads faster than other types. 

In 2024, AI biotech company Owkin partnered with pharma giant AstraZeneca to develop an AI-powered tool designed to pre-screen for gBRCA mutations (gBRCAm) in breast cancer directly from digitized pathology slides. 

The aim of the tool is to speed up and increase access to gBRCS testing that some patients may not be considered for.

That same year, Lunit, a provider of AI-powered solutions for cancer diagnostics and therapeutics and Volpara Health, a company offering AI-powered software to help providers better understand cancer risk, joined forces to develop a comprehensive ecosystem for early cancer detection, cancer risk prediction and independent AI to improve clinical workflows.

In May of that year, Lunit acquired Volpara and integrated its AI breast health platforms, including its Scorecard breast density assessment tool, into its line of AI tools for breast cancer detection.

Before it acquired Volpara, Lunit partnered with one of the country’s biggest private healthcare providers to help raise Sweden’s cancer screening capability. 

In 2023, Lunit signed a three-year agreement with Capio S:t Göran Hospital to supply and license its AI-powered mammography analysis software Lunit INSIGHT MMG. The AI tool enabled the hospital to analyze breast images of approximately 78,000 patients yearly.

According to a study published in the European Journal of Cancer, the fairness and equity of datasets for AI-driven mammogram interpretation might be jeopardized by the underrepresentation of racial and ethnic diversity.

While AI shows promise for enhancing how mammograms are interpreted, particularly in areas where resources are limited, the study’s authors found warning signs regarding the diversity of datasets and the representation of researchers in AI model development, which they said could  “affect the models’ generalizability, fairness and equity.”

For the study, researchers conducted a scientometric review of studies published in 2017, 2018, 2022 and 2023 utilizing screening or diagnostic mammograms for breast cancer detection to “train or validate AI algorithms.”

Of the 5,774 studies identified, 264 met the inclusion criteria. 

“The number of studies increased from 28 in 2017 to 2018 to 115 in 2022 to 2023–a 311% increase. Despite this growth, only 0–25 % of studies reported race/ethnicity, with most patients identified as Caucasian,” the study’s authors wrote. 

“Moreover, nearly all patient cohorts originated from high-income countries, with no studies from low-income settings. Author affiliations were predominantly from high-income regions and gender imbalance was observed among first and last authors.”

The authors concluded: “The lack of racial, ethnic and geographic diversity in both datasets and researcher representation could undermine the generalizability and fairness of AI-based mammogram interpretation.”

Furthermore, recognizing the disparities through diverse dataset collection and comprehensive international collaborations is crucial to guaranteeing fair advancements  in breast cancer care.

Study data revealed that algorithms focusing mostly on caucasian populations could result in inaccurate outcomes and the wrong diagnosis in underrepresented populations. Additionally, patient outcomes may be threatened and current disparities could worsen. 

“The fairness of these AI tools is called into question, as they risk systematically dis-advantaging certain racial, ethnic or socio-demographic groups. To mitigate these issues and ensure that the benefits of AI in BC imaging are equitably distributed, it is essential to prioritize diversity in dataset collection, foster international collaborations that include researchers from lower and middle-income countries and actively incorporate diverse populations in clinical research,” the study’s authors wrote. 

THE LARGER TREND

In February, Google partnered with the Institute of Women’s Cancers, founded by France’s cancer research and treatment center, Institut Curie, to study how AI tools can help address cancer, share science-based health information and support postdoctoral researchers with funding. 

The two entities looked into how AI-based tools can help forecast the progression of cancer and the likelihood of relapse for patients, with the goal of developing more accurate and successful treatments.

The researchers focused on hard to treat women’s cancers, including triple-negative breast cancer, an aggressive type of breast cancer that grows and spreads faster than other types. 

In 2024, AI biotech company Owkin partnered with pharma giant AstraZeneca to develop an AI-powered tool designed to pre-screen for gBRCA mutations (gBRCAm) in breast cancer directly from digitized pathology slides. 

The aim of the tool is to speed up and increase access to gBRCS testing that some patients may not be considered for.

That same year, Lunit, a provider of AI-powered solutions for cancer diagnostics and therapeutics and Volpara Health, a company offering AI-powered software to help providers better understand cancer risk, joined forces to develop a comprehensive ecosystem for early cancer detection, cancer risk prediction and independent AI to improve clinical workflows.

In May of that year, Lunit acquired Volpara and integrated its AI breast health platforms, including its Scorecard breast density assessment tool, into its line of AI tools for breast cancer detection.

Before it acquired Volpara, Lunit partnered with one of the country’s biggest private healthcare providers to help raise Sweden’s cancer screening capability. 

In 2023, Lunit signed a three-year agreement with Capio S:t Göran Hospital to supply and license its AI-powered mammography analysis software Lunit INSIGHT MMG. The AI tool enabled the hospital to analyze breast images of approximately 78,000 patients yearly.