A new study by researchers at Stanford University School of Medicine has revealed a new artificial intelligence model that is more than 90% successful in determining whether brain activity scans are coming from women or men.
The findings, published Feb. 19 in the Proceedings of the National Academy of Sciences, help resolve the long-standing controversy over whether there are reliable sex differences in the human brain and suggest that understanding these differences may be critical to addressing neuropsychiatric disorders that affect women and men differently.
A key motivation for this research is that sex plays a crucial role in human brain development, aging, and the manifestation of psychiatric and neurological disorders," said Vinod Menon, Ph.D., professor of psychiatry and behavioral sciences and director of the Cognitive and Systems Neuroscience Laboratory at Stanford University.
Identifying consistent and reproducible sex differences in healthy** brains is a critical step in gaining a deeper understanding of sex-specific vulnerability in psychiatric and neurological disorders.
Menon is the senior author of the study. The lead authors are Dr. Srikanth Ryali, Senior Research Scientist, and Dr. Yuan Zhang, Academic Staff Researcher.
The "hot spots" that best help the model distinguish between the male and female brains include the default pattern network, a brain system that helps us process self-referential information, and the striatum and limbic networks, which are involved in learning and how we respond to rewards.
The researchers noted that the work did not weigh whether gender-related differences occur early in life or may be driven by hormonal differences or different social circumstances in which men and women may be more likely to encounter.
Revealing brain differences
The extent to which a person's gender affects how their brain is organized and functioning has long been a point of contention among scientists. While we know that our innate sex chromosomes help determine the hormone mixture our brains are exposed to — especially during early development, adolescence, and aging — researchers have long struggled to link sex to specific differences in the human brain.
Male and female brain structures tend to look similar, and previous studies that have studied how brain regions work together have largely failed to find consistent brain sex indicators.
In their current research, Menon and his team have leveraged the latest advances in artificial intelligence, as well as access to multiple large datasets, to conduct more powerful analyses than ever before.
First, they created a way to learn to classify brain imaging data: When researchers showed a brain scan to the model and told it that it was looking at a male or female brain, the model began to "notice" which subtle patterns could help it distinguish.
The model showed better performance compared to previous studies, in part because it used a deep neural network that analyzed dynamic MRI scans. This approach captures the intricate interactions between different brain regions. When the researchers tested the model in about 1,500 brain scans, it almost always told whether the scans were from women or men.
The success of this model suggests that detectable sex differences do exist in the brain, but have not been reliably detected before. The fact that it works well in different datasets, including brain scans from multiple locations in the United States and Europe, makes these findings particularly compelling, as it controls for many of the confounding factors that may plague such studies.
This is very strong evidence that gender is a powerful determinant of human brain organization," Menon said.
Conduct**
Until recently, models like those employed by Menon's team could help researchers classify brains into different groups, but would not provide information about how classification occurred. Today, however, researchers have access to a tool called "explainable AI," which can sift through large amounts of data to explain how the model's decisions are made.
Using explainable AI, Menon and his team identified brain networks that were most important for models to determine whether brain scans were male or female. They found that the model was most often called using the default mode network, striatum, and edge network.
The team then wondered if they could create another model that could use different functional brain traits for women and men to ** participants' performance on certain cognitive tasks. They developed gender-specific models of cognitive ability: one model effectively demonstrated the cognitive performance of men, but not women, and another model effectively demonstrated cognitive performance in women, but not men.
The findings suggest that different functional brain traits of different sexes have significant behavioral effects.
These models worked really well because we were able to separate brain patterns between the sexes," Menon said. "This tells me that ignoring sex differences in brain tissue may cause us to miss out on key factors in neuropsychiatric disorders.
While the team applied their deep neural network model to questions about gender differences, Menon said the model could be used to answer questions about how any aspect of brain connectivity relates to any type of cognitive ability or behavior. He and his team plan to make their model publicly available for use by any researcher.
Our AI models have a very wide applicability," said Menon. "For example, researchers can use our model to look for brain differences associated with learning disabilities or differences in social functioning – aspects that we aspire to better understand to help individuals adapt and overcome these challenges.