It’s for the Global Good — Responsible A.I. Can Support Breast Health Around the World

Nov 17, 2021

Artificial intelligence can open the door to enhanced precision health with data-driven clinical decision support and automated insights for clinical care teams. A.I. is particularly suited to process the massive volumes of healthcare data that is generated routinely, and when integrated into healthcare systems it has the potential to promote better quality and lower-cost healthcare delivery for improved outcomes.

However, A.I. can also spark questions related to biased data, accountability, and privacy. The Responsible A.I. framework focuses on ensuring ethical, transparent, and accountable use of A.I. in a way that is consistent with user expectations, organizational values, and societal laws and norms.

When implemented responsibly, A.I. can support population health management and enable healthcare systems to do more with less.

Global Breast Health Disparities

The ever-widening gap in global economic inequality disproportionately affects under-resourced countries that entirely lack or have insufficient radiology. Diagnostic imaging is extensively relied upon in clinical decision-making such that the implications of inadequate imaging can extend well into healthcare management and patient outcomes.

In low- and middle-income countries, many women are diagnosed at a much later stage of breast cancer due to barriers in accessing appropriate imaging and screening services. As a result, a breast cancer diagnosis is more likely to be fatal. The consequences of such access barriers can be seen in the later-stage cancer diagnosis and breast cancer mortality rates. For example, the age-adjusted death rate in Guyana for 2018 was 24.36 per 100,000 women, compared to 17.49 in the United States.

In high-income countries, large populations of medically underserved women suffer from higher rates of mortality from breast cancer compared to the general population. For example, the consequences of such disparities can be seen in Washington DC. Black women in Washington D.C experience a 50% higher death rate from cancer compared to their white counterparts, and 30% higher than the national average. In general, there are lower coverage rates for supplemental imaging services such as MRI for Medicaid patients compared to private insurance, and looking at the distribution of Medicaid coverage rates by race/ethnicity, 34.3% of those covered by Medicaid in Washington DC are Black citizens compared to 17.7% who are white. Furthermore, outcome disparities may also be exacerbated by the fact that Medicaid patients are more likely to have other challenges, such as lack of transportation, lack of social support, limited financial resources, and comorbidities that contribute to more advanced disease, inadequate treatment and, ultimately, worse survival rates.

Even in well-resourced, non-marginalized populations, the availability of services may be limited due to widespread under-staffed and under-resourced conditions in health systems.

COVID-19 has only exacerbated pre-existing resource challenges and the emergence of new variants continues to further widen the gap in health inequities globally.

How Can A.I. Improve Global Breast Health?

Many breast screening facilities, in high-income and low-income regions alike, are experiencing resource shortages that impact their ability to deliver high-quality breast cancer screening services.

Radiologist shortages are continuing to rise globally, with even greater breast imaging subspecialist shortages. Under-resourced regions in low- and middle-income countries may experience substantially higher clinical staffing shortages than other parts of the world: many have an insufficient number of breast radiologists to run an effective breast cancer screening program. For example, in Kenya, ~4-5 million women need annual mammography, but there are only three fellowship-trained breast imagers in the country.

In the absence of resources, what role can A.I. really play?

Sustainable breast cancer screening service delivery requires efficient patient and process management. The ability to effectively detect breast cancers is predicated on mammography quality control which is a resource-intensive process that requires sufficient numbers of trained radiologists and technologists/radiographers.

While A.I. alone cannot solve these problems, when applied appropriately it has the potential to make a significant impact and support sustainable screening services.

Using A.I. for mammography quality assurance and process automation can improve workflow efficiencies, mitigating the challenges associated with limited numbers of breast radiologists and technologists/radiographers.This allows the clinical team to focus more on moving patients through the breast screening funnel, and less on time-consuming, routine, and easily automated tasks.

Furthermore, access to high-quality mammography technologist education programs is limited in many low- and middle-income countries. Implementation of A.I. solutions supports synchronous learning through automatic assessment of image quality at point-of-care and on-demand, actionable feedback for immediate positioning technique improvements.

This enables technologists to self-correct and empowers them with self-directed learning opportunities, mitigating the challenges associated with limited access to high quality educational programs and resources.. Additionally, mammography-specific workflows and advanced analytics provide a framework for comprehensive and continuous quality assurance processes to improve efficiency and performance.

Combatting Breast Health Disparities — Scalable Technology to Support Health Equity

The fundamental principle of health equity is that all individuals have the fair opportunity to reach their fullest health potential. Achieving this requires societal efforts to address existing inequalities, injustices, and healthcare disparities.

In the context of breast cancer screening, health equity refers to the ability of all women to access high-quality breast imaging regardless of social and/or environmental factors.

The ability of A.I. to support health equity is closely tied to technology and process scalability and relies on the following principles:

1. To help reduce healthcare disparities, A.I. solutions must retain effectiveness and provide the same quality results in all sizes of health systems and different population demographics.

There are many cautionary tales about how A.I. needs to be rigorously validated in the field; A.I. models can be quite susceptible to biases introduced by systematically skewed training data. But good study design and internal and external validation should root out bias. This includes properly curated data and appropriate sampling to establish the training data set.

For example, modeling breast cancer risk on a training data set composed predominantly of white women who are highly educated and of high socio-economic status, then applying the trained model to different patient populations with differing ethnic composition, education and/or socio-economic status can result in poor predictive performance. This may be because the prevalence of breast cancer can be different between the different populations, because the risk factors included in the training data have better data integrity in the different populations, or because women from different socio-economic, cultural, and ethnic groups do not have access to or seek out medical care equivalently.

2. If a technology is not scalable, it may only be able to be applied to large health systems and/or those in high-income regions and thus inherently perpetuates healthcare disparities.

A.I. has the potential to improve global health equity by eliminating the subjectivity and variability associated with visual assessments of measures that are known to impact patient outcomes such as breast density, breast cancer risk and clinical image quality. Using mammographic image data alone to compute these measures has the potential to improve patient care with standardized and complete data collection and improved data integrity.

However, if a solution is not scalable, it cannot be applied population-wide, and so it would only benefit the health outcomes for those who have access. One example may be if A.I. could only be deployed in large, urban, well-resourced health systems, then only those with higher socio-economic status and urban populations would have access. This could be due to a variety of reasons, including the affordability of the solution for the health system, the technological infrastructure required to support the solution, or a high demand on personnel resources that are simply not feasible.

3. If a process is not scalable, it cannot be applied population-wide and thus inherently perpetuates healthcare disparities.

As technological advancements are made, it is important to consider the opportunities that such advancements provide in existing healthcare practices. Breast cancer risk models, for example, used to risk-stratify patients for tailored follow-up breast screening protocols have the potential to improve patient and service delivery in the context of under-resourced regions. But we must consider how this truly works in practice. What is the scalability of this process?

To risk-stratify patients using a standardized breast cancer risk measure, you need complete data capture of risk factors for every woman. On every mammogram. Everywhere in the population. Traditional breast cancer risk models require extensive patient clinical and family history, something that is simply neither practical nor pragmatic at the population level.

This is where the convergence of A.I. automation and healthcare delivery can really excel and deliver scalable solutions for population health management.

We know that A.I. provides the opportunity to improve breast health care delivery, so why isn’t it being used more broadly?

Deployment of A.I. on a global scale has significant challenges — many low- and middle-income countries lack the technology infrastructure and associated digital skills necessary to benefit from A.I. opportunities.

To overcome the challenges of global A.I. adoption, we must first understand them. They can include, but are not limited to the following:

Inadequate and outdated equipment and IT infrastructure (e.g. unreliable internet connectivity)
Shortage of the associated digital skills with new technologies and infrastructure
Shortage of skilled personnel (e.g. technologists, radiologists) in breast imaging
Challenges in the implementation of screening and diagnostic workflows
A lack of breast health navigation to help patients find, access, and navigate complex care pathways.
Differences in perception of and experience with A.I. across cultures

While implementation of A.I. in clinical settings can improve the sustainability of radiology services such as breast cancer screening, a participatory approach is necessary to ensure successful adoption.

Taking a participatory approach to the implementation of A.I. in clinical settings can improve sustainability. This includes multi-stakeholder engagement with A.I. solutions vendors, aid group(s), community stakeholders, and the general public that will be impacted. Overall, this contributes to capacity-building that focuses on empowerment, boosts the trustworthiness of the A.I., and improves the success of adoption.

For example, RAD-AID International is one organization that has focused on connecting radiologically under-served communities globally with medical technology companies in the diagnostic imaging world. To facilitate successful and sustainable radiology improvements, they have developed the Radiology-Readiness Assessment: a framework for implementing radiology in under-resourced regions whereby teams evaluate existing radiology infrastructure and plan an optimized strategy that meets the healthcare needs of specific communities. Analyzing factors related to imaging ensures the strategy can operate within the resource constraints and clinical context of the hospital or community.

Densitas’ Commitment to Responsible A.I.

Responsible A.I. adoption in low- and middle-income countries can often be characterized by a lack of operational partnerships between A.I. developers and aid experts who are bridging the gap between the health system and the technology.

Our partnership with RAD-AID International, an organization whose mission is to improve access to radiology services and quality of care in underserved populations, is a core element of our commitment to improving breast health in under-resourced communities through the development and implementation of responsible A.I.

The aim of our partnership is to provide low-resourced institutions with education and clinical support to enable adoption of sustainable mammography practices by leveraging A.I.-powered technologies. For breast imaging departments in participating institutions, the program is expected to help advance the quality of patient management and support clinical decision-making with the ultimate aim of improving patient outcomes.

Densitas’ scalable and standardized A.I. results generated for every mammogram taken across a health system promote the same quality of care irrespective of linguistic, cultural, and socio-economic barriers.