What is it about?
Mental health disorders are defined and diagnosed using observed behavioral signs and inferred emotional and cognitive symptoms. Many disorders, despite being thought of as separate categories, often have overlapping features, in signs, symptoms, or neural activity. Furthermore, we have yet to identify biomarkers (unique patterns) for any mental health disorder. This makes it challenging to understand how the signs and symptoms used to diagnose people align with neural activity, which means the latter cannot be used for diagnosis. Machine learning has been used in research to classify mental health disorders using only brain imaging data, specifically data showing how different brain regions are connected. Such studies have attempted to classify people with and without a disorder as accurately as possible. However, this approach has not been able to explain how or why disorders are so similar. Our study took a different approach, focusing on the mistakes a brain-based machine learning classification model makes when trying to classify people who have disorders that are similar to each other. We anticipated that disorders that have more symptoms in common, such as schizophrenia and bipolar disorder, would be misclassified more frequently. We then examined how factors such as symptoms and brain networks might help explain why the mistakes were made, which tells us more about how they are similar.
Featured Image
Photo by Growtika on Unsplash
Why is it important?
Focusing on the mistakes made by the model is an approach that has not been undertaken in the context of mental disorders but can provide valuable insight into overlapping disorder characteristics. Further, this approach distinguished between disorders with high similarity, a more informative approach for understanding disorder overlap than attempting to distinguish between individuals with and without a disorder. Such comparisons can identify subtler distinctions that may help improve how mental health disorders are characterized and diagnosed.
Perspectives
This project was particularly exciting to work on because it brings recently developed computational methods into the field of clinical psychology, where they have not often been used. Paying attention to the mistakes that a model makes when classifying disorders based on brain data may advance the long-term goal of including neural signatures in how we characterize disorders. We hope that this article inspires other researchers to try similar machine-learning approaches in their work.
Sarah Olshan
University of Illinois at Urbana-Champaign
Read the Original
This page is a summary of: Making the most of errors: Utilizing erroneous classifications generated by machine-learning models of neuroimaging data to capture disorder heterogeneity., Journal of Psychopathology and Clinical Science, November 2024, American Psychological Association (APA),
DOI: 10.1037/abn0000943.
You can read the full text:
Contributors
The following have contributed to this page
