On the fourth and final day of the congress, the program featured several highly relevant sessions that focused on novel digital diagnostic health technologies in psychiatry, The anhedonic brain, and topics like Speech analysis for the recognition of psychosis.
Here are the highlights from the fourth day in detail:
Towards diagnostic digital health technologies in psychiatry
Transdiagnostic approaches in psychiatry
Speaker: Alexandra Korda
Dr. Alexandra started by referencing a review paper by Professor Paolo Fusar-Poli, which outlines the origins of these approaches, particularly in relation to anxiety, depressive, and eating disorders. Traditionally, these disorders have been treated using specific therapeutic manuals, but this method has limitations, especially when dealing with comorbidities. Transdiagnostic research aims to address these shortcomings by moving away from disorder-specific treatments to a broader framework that targets underlying shared mechanisms across various disorders.
The speaker explained that transdiagnostic approaches represent a significant paradigm shift in how mental disorders are classified, treated, and prevented. Although the potential for a radical shift in clinical care remains uncertain, interest in this field has grown rapidly over the past decade, as evidenced by a sharp increase in publications. She illustrated this with a study comparing schizophrenia, bipolar disorder, and major depression, which found disorder-specific biological alterations but also significant overlaps between schizophrenia and bipolar disorder.
She then delved into the different models of transdiagnostic approaches. The simplest form, “across diagnosis,” compares various diagnostic categories, such as those found in the ICD or DSM systems. More advanced approaches introduce new diagnostic constructs based on clinical or biological phenotypes, aiming to move beyond traditional diagnostic systems and explore new ways of understanding mental health.
Other theme mentioned was the early detection and intervention in mental health. Dr. Alexandra emphasized that early intervention is key to reducing untreated psychosis, improving functional outcomes, and preventing the negative consequences of mental illness, such as social isolation, substance abuse, and poor academic or work performance. She gave an example of the Baltic Early Treatment Services (BeATS) in Germany, which offers customized care and is part of several national and international research programs.
Finally, she highlighted the importance of further developing transdiagnostic therapeutic methods. A notable example is the Unified Transdiagnostic Treatment, which applies the same therapeutic approach across different disorders. These interventions, such as the unified protocol for emotional disorders, are designed to address common mechanisms shared across mental illnesses, particularly targeting comorbidities. She also presented recommendations for future research, including transparent methodologies, clear definitions of diagnostic standards, and the need for external validation studies to ensure generalizability.
Facial expressions analysis for diagnosis
Speaker: Alexandra Korda
In this presentation, Dr. Alexandra explores the innovative use of facial expressions and biomarkers for early psychosis detection, highlighting their recent work. She discussed how video interviews from subjects with various diagnoses were analyzed, marking the first known use of facial expressions to predict depressive symptom severity.
The speaker emphasized that early detection and treatment of mental disorders significantly improve outcomes, yet only a small fraction of patients access early intervention services. Traditional diagnostic methods focus on specific disorders like depression or psychosis, despite symptoms often being non-specific in early stages. This requires a transdiagnostic approach, focusing on symptom dimensions and underlying mechanisms rather than specific diagnoses.
She recorded real-time data from clinical visits, which are rich in information but often underutilized in decision-making. Clinicians’ treatment decisions frequently rely on trial and error, posing risks and increasing costs. Facial recognition technologies, however, can accurately measure patient expressions, providing dynamic symptom tracking and predictive insights.
The study involved clinical semi-structured interviews and self-assessments for depressive symptoms using the BDI scale. They analyzed 32 subjects (58 videos), including follow-ups, with a median BDI score of 18.4. Facial expressions were categorized into action units corresponding to specific muscle movements using the Facial Action Coding System.
Preliminary results showed significant correlations between certain facial action units and depressive symptoms. Wavelet transformations were applied to capture dynamic patterns in these action units, revealing more detailed associations with depressive symptoms. A notable finding was that short-duration, sharp signals in facial expressions correlated strongly with depressive symptoms, aligning with previous research.
Dr. Alexandra also compared depressive patients with those having other diagnoses, using action units in both time and frequency domains. The frequency domain analysis revealed significant differences, highlighting the potential of advanced methods to uncover deeper insights.
Using nonlinear autoregressive neural networks, they attempted to predict future action unit intensities. The prediction accuracy was higher for subjects with lower BDI scores. While a 72% accuracy rate might seem insufficient, it shows promise for real-time analysis and symptom scoring. Future work aims to embed physics into neural networks to better understand facial movements and enhance predictive accuracy.
In conclusion, the study demonstrated that facial expressions are significantly associated with depressive symptom severity. Combining facial analysis with advanced AI tools can facilitate early detection, personalized treatment, and improved patient outcomes. The goal is to develop digital twins, simulations of patients’ mental states using real-time data for personalized predictions and treatments. This represents the future direction of psychiatry, integrating digital health technologies for dynamic and individualized patient care.
Speech analysis for the recognition of psychosis
Speaker: Paolo Brambilla
Dr. Paolo presented a discussion on speech analysis as a potential biomarker for recognizing psychosis. He began by acknowledging the presence of linguistic deficits in psychosis, particularly in schizophrenia, and emphasized that these deficits could be considered core symptoms. He outlined that language is a complex dimension that can be separated into phonology, morphology, syntax, and semantics, with specific brain regions supporting these dimensions. Over the last 20 years, his group has focused on various areas of language, including nonverbal, verbal, literal, and non-literal language, investigating micro and macro dimensions within these categories.
Dr. Paolo highlighted a study showing that psychotic patients often exhibit thought disorders, such as concrete thinking, which affects their comprehension of metaphors and idioms. This impairment was confirmed in patients with schizophrenia when compared to healthy controls. Further investigations using machine learning revealed that linguistic features were more predictive of psychosis than traditional cognitive dimensions like IQ and memory span tests.
He then discussed the differences in linguistic production and comprehension between schizophrenia and bipolar disorder patients. Schizophrenia patients showed impairments in both micro and macro linguistic dimensions, whereas bipolar patients only exhibited reduced speech rates. An Italian-developed grammar comprehension test showed that both bipolar and schizophrenia patients had syntactic comprehension impairments.
Dr. Paolo also explored the first episode psychosis cohort, analyzing non-effective and effective dimensions using the same linguistic tests applied to chronic patients. Results showed that first episode psychotic patients had impairments in both emotional and linguistic prosody, correlating with negative symptoms. These findings suggest that prosody could be an essential aspect of nonverbal communication in psychosis.
The presentation included a pilot study using MRI to assess grammatical comprehension, revealing specific brain activations related to different language tasks. Machine learning applications on speech production in first episode patients indicated that language variables had a higher predictive power for psychosis than cognitive measures.
Dr. Paolo concluded with a discussion on emotional imagery, showing differences in brain activation between healthy individuals, adolescents, and patients with anxiety disorders. Emotional imagery tasks revealed specific brain regions involved in processing emotional versus motor imagery, highlighting the impact of age and gender on these processes.
Novel digital diagnostic health technologies in psychiatry
Speaker: F. Gerrik Verhees
Dr. Verhess, an expert in digital diagnostic health technologies in psychiatry, presented his research on using artificial intelligence to analyze speech for clinical outcomes. He began by appreciating previous speakers and introduced a thought experiment about a “robodoc” that could assist in clinical practice by identifying red flags in patient care. He shared a personal story of a clinical failure where a patient’s deteriorating condition went unnoticed, resulting in a suicide attempt. This highlighted the potential value of AI in improving patient safety.
The core of his research involved using large language models (LLMs) to analyze unstructured data from electronic health records (EHRs). Initially, they faced language barriers with traditional algorithms but found that LLMs, particularly Meta’s LAMA 2 model, could overcome these barriers due to their language-agnostic nature. They collaborated with a lab experienced in LLMs and applied these models to their EHRs to detect signs of suicidality. The results were promising, with the best models achieving over 87% accuracy in identifying cases.
The speaker emphasized the importance of privacy and reliability in using such models, ensuring all data processing remained local to avoid third-party data sharing. They tested the models on a sample of 100 patients and found that the best model showed high specificity and sensitivity. His work was set to be published in the British Journal of Psychiatry.
He acknowledged the evolving landscape of language models, noting the recent release of LAMA 3. While their current findings were promising, he stressed the need for further validation with larger datasets and external validation. The goal is to use AI not only to extract valuable information from unstructured text but also to support clinical practice and advance precision psychiatry.
In summary, Dr. Verhess research showcases the potential of AI and LLMs in enhancing psychiatric care by analyzing unstructured EHR data, ensuring patient safety, and maintaining privacy, with a future vision of integrating AI into everyday clinical practice for better patient outcomes.
Defining treatment-resistant depression for clinical trials
Speaker: Luca Sforzin
In this interactive class the primary focus was on defining TRD clearly to address inconsistencies in clinical research. Dr. Luca introduced the topic by emphasizing that despite numerous treatments for depression, inadequate responses are common, with around half of patients not responding to first-line treatments. The conversation explored how TRD exists on a continuum, from partial to non-responses, making it essential to establish clear definitions, especially for clinical trials. A consensus study, involving 61 experts, created recommendations to standardize the definition of TRD. These guidelines specified that TRD should be defined as the failure to respond to at least two different depression treatments with adequate doses and durations.
The class further addressed the challenges of defining patient groups for research, noting that different mechanisms of action and treatment responses complicate the ability to compare results across studies. Discussions covered how research in major depressive disorder should be multidimensional and focus on biological mechanisms underlying different patient responses. The course also acknowledged regulatory efforts by the FDA and EMA to create standardized definitions and emphasized the need for further research to address remaining gaps, particularly regarding the inclusion of psychotherapy and the variability in patient treatment responses.
Audience questions raised concerns about the difficulty of applying TRD definitions in real-world clinical settings, especially when patient histories are not well-documented. Participants debated the line between partial and full non-response and the role of time and treatment failure in the classification of TRD. The conversation highlighted the importance of reporting specific patient histories, including the type and number of treatments tried, to ensure that studies can accurately identify TRD.
In summary, the session underscored the complexities involved in defining and treating TRD, the need for precise guidelines to improve research outcomes, and the ongoing challenge of balancing standardization with the heterogeneous nature of depression and its treatments.
The anhedonic brain
Speaker: Ciara McCabe
In this lecture, Dr. Ciara discussed anhedonia, a critical symptom of depression that is characterized by the inability to experience pleasure or interest in activities. She began by highlighting the significance of this symptom, often underappreciated in depression, where sadness typically takes center stage. However, it is possible to have depression without sadness, as anhedonia can be a standalone diagnostic criterion. The speaker shared that anhedonia does not respond well to traditional treatments like antidepressants, making it a challenging condition to address.
She touched on the complexity of anhedonia, not merely a lack of pleasure but also a decrease in motivation and interest, as reflected in the DSM’s updated definition. Dr. Ciara emphasized that anhedonia is a potential biomarker for mental health issues and may even predate disorders like depression, persisting in individuals who have recovered from depressive episodes. It also has transdiagnostic relevance, appearing in various conditions such as Parkinson’s disease and substance abuse disorders.
Throughout the session, Dr. Ciara stressed the challenges in defining, measuring, and treating anhedonia. She pointed out the inconsistencies in how symptoms are quantified across different studies and fields of psychology. There is often overlap with other conditions, such as apathy, making it difficult to differentiate and accurately measure anhedonia.
The speaker also noted that research in this area is expanding, especially with younger populations at risk for depression. Her team developed a new questionnaire specifically for adolescents to better capture their experience with anhedonia. In their studies, they found that young people struggle with a sense of identity and disconnection, contributing to their feelings of anhedonia.
The lecture concluded with a discussion on the neurobiological aspects of anhedonia, focusing on reward processing and the involvement of brain regions such as the prefrontal cortex. Dr. Ciara also mentioned the potential role of neurotransmitters like dopamine and serotonin, with new treatments like ketamine showing promise in addressing anhedonia. Overall, the talk underscored the need for more targeted and effective interventions to treat this debilitating symptom.
References
1. FUSAR-POLI, P. Towards diagnostic digital health technologies in psychiatry. In: 37th ECNP Congress, Milan, Italy.
2. SFORZIN, L. Defining treatment-resistant depression for clinical trials. In: 37th ECNP Congress, Milan, Italy.
3. McCABE, C. The anhedonic brain. In: 37th ECNP Congress, Milan, Italy.
Developed under the direction and sponsorship of Johnson & Johnson