Accuracy of retinal image analysis by artificial intelligence in detecting depressive and anxiety syndrome associated with systemic inflammation

• Grant number: 24/13920-5
• Support Opportunities: Regular Research Grants
• Start date: June 01, 2025
• End date: May 31, 2027
• Field of knowledge: Health Sciences - Medicine - Psychiatry
• Principal Investigator: Márcio Antonini Bernik
• Grantee: Márcio Antonini Bernik
• Host Institution: Instituto de Psiquiatria Doutor Antonio Carlos Pacheco e Silva (IPq). Hospital das Clínicas da Faculdade de Medicina da USP (HCFMUSP). Secretaria da Saúde (São Paulo - Estado). São Paulo , SP, Brazil
• Associated researchers: Daniella Mouadeb ; Edecio Cunha Neto ; FABIO MORAES CORREGIARI ; Jochen Kumm ; LUCAS MARQUES GANDARELA ; Pedro Samaia de Castro ; Ricardo Castilho da Silva ; Wagner Farid Gattaz

Abstract

Background: Several studies point to increased inflammatory activity in a significant portion of patients with major depression and anxiety disorders. On the other hand, there is evidence that the disorders as defined in current diagnostic manuals may not accurately describe the depressive-anxious symptom cluster associated with systemic inflammation. Additionally, the measurement of plasma inflammatory markers is costly and requires blood collection. Retinal image analysis by artificial intelligence (AI) may be a cheaper and less invasive alternative. Main Aim: To empirically define a set of depressive and anxious symptoms associated with systemic inflammation and to test the predictive ability of this syndrome from retinal images analyzed by artificial intelligence. Methodology: The study population will consist of internet-screened subjects with at least mild depressive and anxious symptoms and age between 18 and 45 years. Subjects with neurological, somatic or retinal diseases that may interfere with the results will be excluded. Subjects with psychotic conditions, significant substance dependence or abuse (except nicotine), and severe suicidal ideation will also be excluded. Subjects will be included until 200 subjects with complete data will be randomly assigned to a learning cohort (100 subjects) and a validation cohort (100 subjects). All will undergo questionnaires assessing symptoms and diagnoses of mental disorders, dosing of serum markers of inflammation, and AI retinal imaging analysis. From the Learning Cohort, a cluster analysis of data (latent class analysis) will be performed and a syndrome of depressive and anxious symptoms associated with peripheral markers of inflammation (inflammation-associated neuroticism syndrome) will be defined. Using AI, retinal imaging patterns associated with systemic inflammation and the syndromes The validation cohort will allow us to assess the degree of accuracy of retinal analysis in identifying inflammation-associated neuroticism syndrome, i.e. its sensitivity, specificity, positive predictive value and negative predictive value. (AU)