Project

ABSTRACT

Schizophrenia (SZ) affects approximately 24 million people or 1 in 300 people (0.32%) worldwide (WHO, 2022). This rate is 1 in 222 people (0.45%) among adults.

Among the patients in the SZ spectrum, there exist patients whose significant symptoms persist despite adequate antipsychotic treatment, classified as Treatment Resistant Schizophrenia (TRS) patients. TRS is a severe condition affecting almost 30% of schizophrenia patients (Iasevoli et al., 2016). However, TRS is diagnosed very late during the course of the disorder, preventing from switching patients to more effective treatments (i.e., clozapine) and non-pharmacological therapeutic strategies, with enormous individual sufferance and community economic costs. An early and accurate diagnosis of TRS can enable clinicians to propose more suitable pharmacological and non-pharmacological therapies, may improve the Quality of Life (QoL) of TRS patients, and may spare large economical resources.

SPECTRA Project aims at designing a Decision Support System (DSS) based on AI techniques and the use of cutting-edge Technologies for the early diagnosis of TRS patients. The innovative aspect of the Project is to combine typical screening procedures used in standard clinical practice, with ICT-based assessment techniques based on Machine Learning algorithms.

SPECTRA will conduct a field study involving real patients from the Unit for Treatment-Resistant Psychosis, University "Federico II" of Naples. After being categorized as TRS or nonTRS, eligible patients will enter the study. The SPECTRA staff will collect historical clinical patient data, such as Magnetic Resonance Images, questionnaire scores, demographic data, geographical and feeding data together with real-time data collected during the patient’s screening, such as physical data acquired by IoT sensors (ECG, temperature, EEG, etc,), video and speech. Thus, standard data and data acquired by using ICT technologies will be adopted for training a Machine Learning/Deep Learning model for detecting TRS patients. However, one of the main problems with the adoption of AI solutions for supporting decisions in healthcare is that clinicians lack trust in the black-box operation. Thus, another project's main objective is to define interaction models suitable for supporting clinicians during the TRS diagnosis by exploiting eXplainable Artificial Intelligence (XAI) methods, which will explain how AI black-box models generate predictions. XAI techniques will be used to provide clinicians with more precise details about the peculiarities of some subdimensions of SZ in TRS patients, supporting them in choosing a more appropriate alternative pharmacological treatment.

As a secondary outcome, a TSR-nonTRS dataset will be created that will be useful for the scientific community for experimenting with Machine Learning approaches to the diagnosis of TRS in people suffering from schizophrenia.

Architecture