Using support vector machine to explore the difference of function connection between deficit and non-deficit...

Area of Science:

• Neuroscience
• Psychiatry
• Medical Imaging

Background:

• Schizophrenia presents with diverse symptom profiles, broadly categorized into deficient schizophrenia (DS) and non-deficient schizophrenia (NDS) based on negative symptoms.
• Understanding the neural underpinnings of these subtypes is crucial for targeted treatments.
• Limited research exists on functional connectivity (FC) differences between DS and NDS, particularly using gray matter volume (GMV) derived regions of interest (ROIs).

Purpose of the Study:

• To investigate functional connectivity (FC) alterations between DS and NDS subtypes of schizophrenia.
• To identify ROIs using machine learning and differential GMV for FC analysis.
• To explore the relationship between altered FC and clinical symptoms, including thalamic functional connectivity imbalance.

Main Methods:

• Utilized resting-state fMRI data from 16 DS, 31 NDS, and 38 healthy controls (HC).
• Employed a support vector machine (SVM) on GMV data to classify DS and NDS, identifying high-weight regions as ROIs.
• Conducted whole-brain FC analysis, thalamic FC imbalance analysis, and partial correlation with clinical scales (BPRS, SANS, SAPS).

Main Results:

• SVM achieved high classification accuracy between DS and NDS.
• NDS showed increased FC between right inferior parietal lobule (IPL.R) and bilateral thalamus/lingual gyrus, and right inferior temporal gyrus (ITG.R) and Salience Network (SN) compared to HC.
• DS exhibited increased FC between right thalamus (THA.R) and Visual Network (VN), and ITG.R and right superior occipital gyrus versus HC.
• DS demonstrated decreased FC between ITG.R and left superior/middle frontal gyrus compared to NDS.
• Both subtypes displayed thalamic FC imbalance: frontotemporal-THA.R hypoconnectivity and sensory motor network (SMN)-THA.R hyperconnectivity.
• THA.R-SMN FC negatively correlated with SANS scores in DS and positively with SAPS scores in NDS.

Conclusions:

• Machine learning-based ROI identification from GMV successfully highlighted FC differences between DS and NDS.
• Findings provide novel insights into the neural pathophysiology of schizophrenia subtypes at local and network levels.
• The study underscores the utility of integrating neuroimaging and machine learning for understanding complex brain disorders.