Scientists from Russia and Bulgaria have developed a computational system based on contrastive learning that distinguishes patients with major depressive disorder from healthy individuals with 86% accuracy. The system uses two algorithms to identify differences in brain MRI scans. The technology could enable more accurate diagnosis of clinical depression at early stages, when behavioural changes are just beginning to emerge, the researchers said.
First, the algorithm selected features of brain network organisation that best reflected behavioural changes – such as alterations in frontal cortex connectivity. The researchers then applied a contrastive approach: the programme sought differences between depressed patients and healthy individuals based on those selected features, while ignoring variations within each group. Unlike conventional methods, this approach detected small but clinically significant brain structure differences distinguishing patients from controls.
The researchers achieved 86% accuracy in distinguishing healthy individuals from those with depression.
By comparison, traditional diagnostic methods involving clinical assessment showed accuracy of around 50% – close to random chance.
The scientists also confirmed that major depressive disorder is not a focal pathology but involves dysfunction across multiple neural networks. The proposed approach identified 20 key connections involved in depression, whereas traditional methods detected only five.
“The proposed algorithm will improve diagnostic accuracy and pave the way for more personalised treatment approaches, helping to enhance patients’ quality of life and reduce the socio‑economic burden of the disease,” said Semyon Kurkin, a project participant and chief researcher at the Institute of Applied Artificial Intelligence and Digital Solutions of Plekhanov Russian University of Economics. “We plan to apply this approach to other psychiatric and neurological conditions, such as schizophrenia and bipolar disorder.”
