Support Vector Machines Based Prediction of Schizophrenia Using Neuroimaging Features

Poster No:

M119 

Submission Type:

Abstract Submission 

Authors:

Yinghan Zhu1, Hironori Nakatani1, Walid Yassin2, Naohiro Okada3, Omasu Abe4, Hidenori Yamasue5, Kiyoto Kasai3, Shinsuke Koike6

Institutions:

1University of Tokyo Center for Evolutionary Cognitive Sciences, Tokyo, Japan, 2University of Tokyo 2Department of Child Neuropsychiatry, Graduate School of Medicine, Tokyo, Japan, 3University of Tokyo Department of Neuropsychiatry, Graduate School of Medicine, Tokyo, Japan, 4University of Tokyo Department of Radiology, Graduate School of Medicine,, Tokyo, Japan, 5Hamamatsu University School of Medicine Department of Psychiatry, Hamamatsu, Japan, 6University of Tokyo Institute for Diversity & Adaptation of Human Mind, Tokyo, Japan

Introduction:

This research aims whether the participants who are ultra-high risk for psychosis (UHR) and first episode of schizophrenia (FEP) would be classified into schizophrenia spectrum, using a machine learning classifier based on T1-weighted brain images for healthy control and chronic schizophrenia. This biological based classification is considered helpful for diagnosis and future prognosis(Lindstrom, Wieselgren, & Vonknorring,1994; McGorry et al., 1995; Norman, Malla, Cortese, & Diaz,1996).

Methods:

Fourty-two patients with chronic schizophrenia, 35 UHR individuals, 22 patients with FEP, and 143 heathy controls were recruited from the Department of Neuropsychiatry, The University of Tokyo Hospital, Japan. There were two different protocols used for data acquisition. Individual T1-weighted images were segmented to produce images of different tissue types (gray matter, white matter, and cerebro-spinal fluid [CSF]) with SPM12 software (www.fil.ion.ucl.ac.uk/spm). We used the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) (Ashburner, 2007) option in the SPM12 toolbox, which creates a study-specific template for segmentations. The tissue classified gray matter and white matter maps were smoothed using a Gaussian smoothing kernel of 8-mm full width at half-maximum (FWHM).
To build a classifier that predict the possible labels for individuals with UHR and FEP, the participants' gray matter images for heathy control, chronic schizophrenia, and autism was used. Before classification, the principal component analysis (PCA) was perfumed for dimension reduction. The classification accuracy was evaluated using leave-one-out cross-validation (LOOCV) with linear support vector machine (SVM) classifier. Best parameters of both PCA and SVM were optimized by grid search(Figure 1). All analyses were conducted using Python 3.6.5.
Supporting Image: GV.png
   ·Fig. 1 The result of Gridsearch
 

Results:

There were 40 components used for classification. The best accuracy of train set was 83.73%, and the accuracy for test set was 84.21%(Figure 2). Eighty-nine percent of patients with UHR are labeled as healthy controls, and 11% of them are labeled as schizophrenia. On the other hand, 20% of patients with FEP are labeled as heathy controls, and 80% of them are labeled as schizophrenia.
Supporting Image: confusionheatmap1png.png
   ·Fig. 2 Confusion matrix of the test set
 

Conclusions:

In this research we demonstrated that using the whole brain voxel information for classification gave us more accurate labels than chance level for the population that was not included in the training process. The results showed that schizophrenia features compared to autism could be contained in brain images for UHR and FEP. Furthermore, it is noteworthy that the percentage of predicted labels as Schizofrenia for UHR and FEP are opposite.

Disorders of the Nervous System:

Schizophrenia and Psychotic Disorders 1

Imaging Methods:

Anatomical MRI 2

Keywords:

Machine Learning
MRI
Schizophrenia
STRUCTURAL MRI
Structures

1|2Indicates the priority used for review

My abstract is being submitted as a Software Demonstration.

No

Please indicate below if your study was a "resting state" or "task-activation” study.

Other

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Patients

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.

Yes

Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Yes

Please indicate which methods were used in your research:

Structural MRI

For human MRI, what field strength scanner do you use?

3.0T

Which processing packages did you use for your study?

SPM

Provide references using author date format

Lindstrom, E.(1994). Interrater reliability of the structured clinical interview for the positive and negative syndrome scale for schizophrenia. Acta Psychiatr. Scand. 89 (3), 192–195.
McGorry, P.D.(1995). Spurious precision: procedural validity of diagnostic assessment in psychotic disorders. Am. J. Psychiatry 152 (2), 220–223.
Norman, R.M.G.(1996). A study of the interrelationship between and comparative interrater reliability of the SAPS, SANS and PANSS. Schizophr. Res. 19 (1), 73–85.