In a recent development, researchers from KAIST have created a technology that uses artificial intelligence (AI) to diagnose depression and evaluate the effectiveness of treatments by analyzing everyday behaviors. On January 13, KAIST announced that Distinguished Professor Won Do Heo’s research team from the Department of Life Sciences developed an AI technology capable of analyzing typical behavior patterns in animal models and demonstrated its ability to detect depressive symptoms varying by gender and severity from everyday actions.
The research team observed that physical movement patterns, such as limb movements, postures, and facial expressions, differ between depressed patients and the general population. Based on these observations, they developed a platform called ‘CLOSER’ (Contrastive Learning-based Observer-free Analysis of Spontaneous Behavior for Ethogram Representation), which can automatically capture subtle behavioral changes associated with depressive states.
By utilizing contrastive learning algorithms, CLOSER breaks down behaviors into small units for analysis, precisely distinguishing minute behavioral changes that are difficult for the human eye to detect. The platform successfully differentiated depressive states based on gender and symptom severity. Post-analysis indicated that stress significantly impacts the frequency and flow of behaviors rather than the capability to perform actions.
In depression models, there were distinct behavioral changes induced by stress that varied by gender. For instance, male mice exhibited reduced exploratory and rotational behaviors, while such behaviors increased in female mice. These behavioral changes became more pronounced with prolonged stress exposure.
Moreover, the research indicated that when depression was induced by persistent stress or inflammation, noticeable changes in daily activities were observed, whereas merely administering the stress hormone corticosterone resulted in negligible behavioral changes. This confirmed that everyday behavior observation alone could distinguish between different depression states based on cause or gender.
Upon administering antidepressants in depression models, the researchers found partial recovery in the behavioral syllables (basic behavioral units) and grammar (flow and patterns of behavior) altered by stress. Different antidepressants restored behaviors differently, allowing researchers to identify a ‘Behavioral Fingerprint’ that distinguishes which medications are more effective.
Distinguished Professor Won Do Heo stated, “This research signifies the initial implementation of a preclinical framework that integrates AI-based everyday behavior analysis platforms for tailored diagnosis and treatment evaluation of depressive disorders, setting a vital groundwork for developing personalized treatments for mental health patients and advancing precision medicine.”
The study, led by Hyunshik Oh, a Ph.D. candidate in the Department of Life Sciences at KAIST as the first author, was published online in the international academic journal ‘Nature Communications’ on December 30th of the previous year.