3 hours ago
The power of artificial intelligence lies in extracting otherwise unidentifiable patterns from large amounts of data. Recent advances in brain imaging and recording technologies have enabled these modalities to produce enough neural and behavioral data to train AI models. And we can use these models to capture neural activity dynamics, quantify an organism’s behavior or map the relationship between brain and behavior, revealing the underlying circuitry of complex processes such as vision, decision-making, learning and memory.
This is the holy grail of neuroscience—to understand how the brain gives rise to our behavior and how our environment and biology affect both. If we can answer these questions, we may be able to not only better treat brain disorders but also understand the essence of what makes us human.
However, AI algorithms are only as good as the data on which they are trained. In human neuroscience, health factors unique to women are dramatically understudied. To date, only 0.5 percent of all neuroscience papers address women’s health issues, and few large-scale neuroimaging datasets include detailed information about women’s health, such as menstrual cycle characteristics, hormonal contraceptive use, past pregnancies, reproductive or endocrine conditions and menopause symptoms. The first human MRI took place in 1977, yet the first continuous imaging study of the brain across pregnancy was not published until 2024. Laura Pritschet and her colleagues’ landmark precision-imaging study, and others that followed, revealed pregnancy to be a time of massive brain change. We need much more work still to understand the longer-term and brain health implications of the sweeping pregnancy-related changes they identified.
Despite these encouraging recent advances, the sheer volume of women’s brain health data that is still missing severely constrains AI’s ability to reveal how life events that women experience shape their brains. If the data do not exist, they cannot be used to robustly and effectively train AI to answer questions about how genetics, sex hormones, gender and events affect or are affected by a woman’s brain. The pace at which AI is progressing means this lack of data will keep women’s brain health in the dark ages. To prevent the field from being further excluded from the AI movement, we need brain data on women—and lots of it—fast.
As co-director of the Artificial Intelligence Core for the Ann S. Bowers Women’s Brain Health Initiative, a research consortium of more than 10 institutes focused on studying women, I am working with my colleagues to fill this gap.
W
Progress since then has been steady, if slow. The long history of exclusion means female-specific data are still lacking, which creates problems for AI tools. AI algorithms trained on small data are often overfitted to the dataset’s esoteric nature and not representative of broader realities. Clever approaches exist to train AI models on small data, but they cannot be a long-term replacement for missing data—particularly when brain measures are so variable across different people and across time within the same person.
To remedy this issue, we need to standardize the reproductive health information we gather alongside new and ongoing neuroimaging studies. At the Women’s Brain Health Initiative, we aim to create the world’s most comprehensive standardized reproductive health questionnaire. Adding this questionnaire to neuroimaging studies will begin to address the lack of women’s data, as a first step to understanding how reproductive health affects the brain.
In addition to collecting better data, we also need more women to join the field of AI. Last year, my AI Core co-director Nina Miolane and I helped organize an AI data challenge centered around women’s brain health. We partnered with Women in Data Science, an advocacy group that holds annual data challenges to get more women involved with the largely male-dominated field of data science. In the United States, only about 20 to 25 percent of computer science university graduates are women; cultural and societal influences are certainly a contributing factor. The community that surrounds the Women in Data Science data challenges creates an accessible and welcoming environment for women who are interested in joining the field of AI.
Designing the data challenge itself was extremely difficult because of the lack of data on women’s brain health. We struggled to find appropriate datasets; only a couple of large-scale, publicly available datasets contain women’s health information, such as past pregnancies, hormonal contraceptive use, menstrual cycle information or menopausal state. In the end, the data challenge focused on investigating sex differences in adolescents with attention-deficit/hyperactivity disorder. It was an enormous success, with more than 16,000 entries from 100 countries around the world.
The AI gender imbalance also exists among tenured neuroscience faculty: Only about 30 percent are women. Given the fact that 50 percent of neuroscience Ph.D. students are women, this dearth demonstrates the so-called “leaky pipeline” in academia. Although being a woman is not a prerequisite for studying women’s brain health, most of those who do are women, presenting another big demographic gap for the application of AI to women’s brain health data. To put it bluntly, we need to recruit more AI researchers to be interested in women’s brain health, and we need to make women’s brain health researchers more fluent in AI. Having more researchers at this intersection will make it possible (once the data are there) to use AI to reveal the mysteries of the human brain and to ensure those discoveries are relevant for all.
