High Classification Accuracy for Schizophrenia with Rest and Task fMRI Data

Wei Du, Vince D. Calhoun, Hualiang Li, Sai Ma, Tom Eichele, Kent A. Kiehl, Godfrey D. Pearlson, Tülay AdalıFrontiers in Human Neuroscience 2012 · DOI

Abstract

This paper presents a novel framework for extracting classification features from functional magnetic resonance imaging (fMRI) data to distinguish healthy controls from patients with schizophrenia. By integrating a two-level feature identification scheme with kernel principal component analysis (KPCA) and Fisher's linear discriminant analysis (FLD), the study achieves high classification rates across both resting-state and task-based datasets. Using leave-one-out cross-validation, features from the default mode network (DMN) alone yielded over 90% accuracy. A majority vote method combining multiple brain networks further increased accuracy to 98% for auditory oddball (AOD) task data and 93% for resting data. These results suggest that specific brain networks, particularly the DMN and temporal regions, serve as robust biomarkers for schizophrenia.

Introduction

Diagnosing schizophrenia has historically relied on clinical observations of symptoms rather than objective biological markers. Functional Magnetic Resonance Imaging (fMRI) provides a non-invasive method to observe brain activity, but the resulting data is high-dimensional and often obscured by noise. Previous attempts to automate diagnosis using fMRI have frequently struggled with the "curse of dimensionality" and the inability of linear models to capture the complex, non-linear interactions within the brain. This study addresses these challenges by proposing a robust, automated pipeline designed to maintain high accuracy across different brain states, including active tasks and resting periods.

Methodology

The proposed framework utilizes a three-phase feature extraction process applied to data decomposed into Independent Components (ICs) via Group ICA.

  1. Two-Level Feature Identification: To prevent overfitting, the researchers applied statistical t-tests. One-sample t-tests identified "active" voxels within groups, while two-sample t-tests identified "discriminative" voxels between patients and controls. Only voxels meeting both criteria were retained.
  2. Kernel PCA (KPCA): Recognizing that brain activity involves non-linear dependencies, the study employed KPCA. This technique uses a "kernel trick" to project data into a higher-dimensional space where complex patterns become linearly separable.
  3. Fisher’s Linear Discriminant (FLD): Finally, FLD was applied to the KPCA features to maximize the separation between the healthy and patient groups while minimizing within-group variance.
Schematic of the feature extraction and cross-validation process for each brain component.

The evaluation used a leave-one-out cross-validation approach. For the final diagnostic call, a majority vote method was implemented, aggregating results from multiple key brain networks.

Datasets

The study involved 56 participants, consisting of 28 healthy controls (HC) and 28 chronic schizophrenia patients (SZ). Each participant was scanned under two conditions:

  • Auditory Oddball (AOD) Task: A task requiring subjects to respond to infrequent target sounds.
  • Resting-State: A condition where subjects rested quietly while focusing on a crosshair.

Results

The quantitative performance of the framework was exceptionally high, demonstrating the effectiveness of non-linear feature extraction:

  • Task-Based Accuracy: The AOD task data reached a peak accuracy of 98% using a majority vote of five components (DMN, temporal, motor-temporal, fronto-insula, and medial visual). This included a sensitivity of 0.98 and a specificity of 1.00.
  • Resting-State Accuracy: The model achieved 93% accuracy using a combination of networks.
  • DMN Performance: Features from the Default Mode Network alone yielded over 90% accuracy in both datasets, highlighting its central role in the pathophysiology of schizophrenia.

Validation tests showed that removing the initial statistical feature-selection stage caused accuracy to drop to approximately 73%, confirming that "cleaning" the data before non-linear transformation is essential for high performance.

Discussion and Limitations

The findings underscore the importance of non-linear analysis in neuroimaging. The slightly higher accuracy in task-based data suggests that cognitive effort may accentuate abnormalities in brain networks that are subtler during rest.

While promising, the study is limited by its sample size (n=56) and the fact that all patients were on psychotropic medications, which may influence brain patterns. Future research should focus on larger, medication-naive cohorts and test the pipeline's consistency across multiple clinical sites to ensure its viability as a real-world diagnostic tool.

Summary updated

Abstract

We present a novel method to extract classification features from functional magnetic resonance imaging (fMRI) data collected at rest or during the performance of a task. By combining a two-level feature identification scheme with kernel principal component analysis (KPCA) and Fisher's linear discriminant analysis (FLD), we achieve high classification rates in discriminating healthy controls from patients with schizophrenia. Experimental results using leave-one-out cross-validation show that features extracted from the default mode network (DMN) lead to a classification accuracy of over 90% in both data sets. Moreover, using a majority vote method that uses multiple features, we achieve a classification accuracy of 98% in auditory oddball (AOD) task and 93% in rest data. Several components, including DMN, temporal, and medial visual regions, are consistently present in the set of features that yield high classification accuracy. The features we have extracted thus show promise to be used as biomarkers for schizophrenia. Results also suggest that there may be different advantages to using resting fMRI data or task fMRI data.
BibTeX
@inproceedings{du2012high,
  author = {Wei Du and Vince D. Calhoun and Hualiang Li and Sai Ma and Tom Eichele and Kent A. Kiehl and Godfrey D. Pearlson and Tülay Adalı},
  title = {{High Classification Accuracy for Schizophrenia with Rest and Task fMRI Data}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2012},
  doi = {10.3389/fnhum.2012.00145},
}
Figure