Artificial Neural Networks for First-episode Schizophrenia Classification Based on MRI Data

VYŠKOVSKÝ, Roman, Eva JANOUŠOVÁ and Daniel SCHWARZ. Artificial Neural Networks for First-episode Schizophrenia Classification Based on MRI Data. In ISCAMI 2016 Book of Abstracts. 2016.

Basic information

• Original name: Artificial Neural Networks for First-episode Schizophrenia Classification Based on MRI Data
• Authors: VYŠKOVSKÝ, Roman, Eva JANOUŠOVÁ and Daniel SCHWARZ.
• Edition: ISCAMI 2016 Book of Abstracts, 2016.

Other information

• Original language: English
• Type of outcome: Conference abstract
• Field of Study: 20200 2.2 Electrical engineering, Electronic engineering, Information engineering
• Country of publisher: Czech Republic
• Confidentiality degree: is not subject to a state or trade secret
• Organization unit: Faculty of Medicine
• Keywords (in Czech): umělé neuronové sítě; diagnostika schizofrenie
• Keywords in English: artificial neural networks; schizophrenia diagnostics
• Tags: International impact

Abstract

Diagnostics of schizophrenia is based on self-report, manifestation of symptoms, psychiatrist's experience and interview with patient's family. No objective diagnostic tool is available and therefore there is a demand for computer-aided tools that would help in recognition of schizophrenia in the first episode. Early detection could accelerate deployment of antipsychotics and improve prognosis of patients suffering from this severe mental disorder. In this study, we applied several types of artificial neural networks - multilayer perceptron, radial basis function network (RBF) and learning vector quantization network - and explored the influence of their parameters such as number of input or hidden neurons on the classification performance. Our dataset consisted of 104 T1-weighted magnetic resonance images (52 patients and 52 control subjects) processed with the use of optimized voxel based morphometry. We selected only the most significant voxels as the features based on two-sample t-test. To ensure a methodologically appropriate framework, both the feature selection step and the classification step were validated using leave-one-out cross-validation. The highest accuracy among the proposed classifiers was 76% and it was achieved by RBF network.