July 12, 2024

expert reaction to study looking at sex, gender, and brain network patterns using brain imaging in children

A study published in Science Advances looks at the association of brain network patterns with children’s birth sex and self/parent-reported gender.

Dr Niall Bourke, Research Fellow, Centre for Neuroimaging Sciences, Institute of Psychiatry, Psychology & Neuroscience (IoPPN), King’s College London, said:

“This important work explores a significant gap in existing academic literature.  Recognising gender as a distinct feature to sex will lead to more precise scientific investigations, improving validity and promote personalised and equitable healthcare.  The authors take care to provide a balanced discussion, including limitations of the study acknowledging factors that may influence gender identity such as geography, social structures, and age.  We need to be careful not to misrepresent these findings in political or philosophical issues.  A detailed discussion framing the context of sex and gender is provided in the supplementary which could be published independently as a commentary and is a recommended read for the scientific community.”

Dr Anne-Lise Goddings, Clinical Academic Consultant in Paediatrics and Adolescent Health at London North West University Healthcare NHS Trust; and Honorary Senior Clinical Lecturer at Imperial College London with a research background in adolescent cognitive neuroscience and neuroimaging, said:

“This study reports relatively weak brain functional connectivity associations with parent-reported ‘gender’ over and above sex assigned at birth.  It’s crucial when interpreting this study to contextualise that this ‘gender’ measure focuses mainly on how much parents report their children engaged in stereotypical ‘gendered’ play.  Parents answered questions including about how much their child plays with “girl-type dolls such as ‘Barbie’”, “boy-type dolls such as ‘GI-Joe’”, “how much they play sports with girls (but not boys)” and “with boys (but not girls)”, and how much they “imitate male and female TV and movie characters”.

“This measure of gender doesn’t capture the broader concept of gender identity which incorporates an individual’s own feelings and perceptions of their identity.

“In the absence of clear hypotheses or accounting for confounders, the findings of the study are of limited impact and should be interpreted with caution.  Later Adolescent Brain Cognitive Development project (ABCD) study waves and other cohort studies may help to improve understanding of this complex topic.”

Prof Derek Hill, Professor of Digital Health, and of Medical Imaging Science, UCL, said:

“This paper seeks to use a special type of brain scan to better understand the brain biology behind sex and gender in children.  This type of MRI brain scan is often used in research studies, and can be analysed with sophisticated algorithms to produce maps of functional connectivity: how different parts of the brain are connected together.

“The paper uses a type of artificial intelligence called machine learning to build a computer model that can predict a child’s sex at birth or self-reported and parent reported gender from their brain scan.  In this way they can determine which type of functional connectivity are associated with sex and gender.

“To test and train this model, they used brain scans from nearly 5000 children aged 9-10 years, for whom the sex assigned at birth and their self-reported and parent-reported gender identity was known.

“This is a relatively large imaging study, using carefully collected data from nearly 5000 children, and state-of-the art brain image analysis.  However, the results must be viewed with caution because:

• artificial intelligence algorithms are very data hungry, and the number of children in some of the sex / gender categories in this dataset is small, meaning reliability of the model may be limited.
• the authors use a method called “cross validation” to train and test their Artificial Intelligence model on this single dataset, which often gives an exaggerated impression of how well the model works.  Very often, Artificial Intelligence models trained and tested in this way suffer from “over-fitting”, which means they appear to perform better on the training data than when applied to other people in a real-world setting.  A more rigorous approach, which is now required by medical device regulators for any algorithm that might be used for diagnosis or management of patients, is to use a truly independent dataset for testing and training.  Without demonstrating that this model performs equally well on an independent dataset, the confidence in the conclusions must be considered low.
• the children looked at are all very similar age – with imaging collected when they are aged 9 or 10 and the gender and self report gender being recorded a year later.  It isn’t at all clear whether the results presented here would translate to children of other ages.”

‘Functional brain networks are associated with both sex and gender in children’ by Elvisha Dhamala et al. was published in Science Advances at 19:00 UK time on Friday 12 July 2024.

DOI: 10.1126/sciadv.adn4202

Declared interests

Dr Niall Bourke: “I do analysis consultancy for https://www.advanced-mri.co.uk/ through my company metacognition.ltd.”

Dr Anne-Lise Goddings: “I don’t think I have any declarations to note, but for full openness, below is my current and previous funding:

Current funding bodies:

MRC

NHS.

Previous funding bodies:

NIHR

MRC

Academy of medical sciences.

I haven’t had any industry funding of my own research, scientific meetings, advisory roles or employment.  I’m not aware of any industry funding to my current or previous department.”

Prof Derek Hill: “No conflicts of interest relevant to this study, but currently CEO of Panoramic Digital Health, and previously co-founder of IXICO.”