Classification of Schizophrenic Traits in Transcriptions of Audio Spectra from Patient Literature: Artificial Intelligence Models Enhanced by Geometric Properties

Abstract

ABSTRACTSchizophrenia is a severe mental illness that affects approximately 1% of the global population and presents significant challenges for patients, families, and healthcare professionals. Characterized by symptoms such as delusions, hallucinations, disorganized speech or behavior, and cognitive impairment, this condition has an early onset and chronic trajectory, making it a debilitating challenge. Schizophrenia also imposes a substantial burden on society, exacerbated by the stigma associated with mental disorders. Technological advancements, such as computerized semantic, linguistic, and acoustic analyses, are revolutionizing the understanding and assessment of communication alterations, a significant aspect in various severe mental illnesses. Early and accurate diagnosis is crucial for improving prognosis and implementing appropriate treatments. In this context, the advancement of Artificial Intelligence (AI) has provided new perspectives for the treatment of schizophrenia, with machine learning techniques and natural language processing allowing a more detailed analysis of clinical, neurological, and behavioral data sets. The present article aims to present a proposal for computational models for the identification of schizophrenic traits in texts. The database used in this article was created with 139 excerpts of patients’ speeches reported in the book “Memories of My Nervous Disease” by German judge Daniel Paul Schreber, classifying them into three categories: 1 - schizophrenic, 2 - with schizophrenic traits and 3 - without any relation to the disorder. Of these speeches, 104 were used for training the models and the others 35 for validation.Three classification models were implemented using features based on geometric properties of graphs (number of vertices, number of cycles, girth, vertex of maximum degree, maximum clique size) and text entropy. Promising results were observed in the classification, with the Decision Tree-based model [1] achieving 100% accuracy, the KNN-k-Nearest Neighbor model observed with 62.8% accuracy, and the ‘centrality-based’ model with 59% precision. The high precision rates, observed when geometric properties are incorporated into Artificial Intelligence Models, suggest that the models can be improved to the point of capturing the language deviation traits that are indicative of schizophrenic disorders. In summary, this study paves the way for significant advances in the use of geometric properties in the field of psychiatry, offering a new data-based approach to the understanding and therapy of schizophrenia.