Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Explainable Deep Learning-Based Method For Schizophrenia Diagnosis Using Generative Data-Augmentation

Mehrshad Saadatinia, Armin Salimi-Badr

TL;DR

This work addresses the lack of objective diagnostic tools for schizophrenia by leveraging EEG-derived spectrograms and a CNN classifier trained with generative data augmentation. It compares VAEs and WGAN-GP for synthetic data, finding that a VAE-based augmentation (VAE-700) substantially improves accuracy to 99.0% on a 16-channel dataset, while the 19-channel dataset already achieves 99.6% accuracy without augmentation. The paper also employs LIME to provide explanations, revealing frequency-region patterns that differentiate schizophrenic and healthy EEG spectra, thereby enhancing trust in the model. Overall, the study demonstrates that targeted generative augmentation can generalize well on EEG-based schizophrenia diagnosis and that explainability methods help interpret neural decisions for clinical use.

Abstract

In this study, we leverage a deep learning-based method for the automatic diagnosis of schizophrenia using EEG brain recordings. This approach utilizes generative data augmentation, a powerful technique that enhances the accuracy of the diagnosis. To enable the utilization of time-frequency features, spectrograms were extracted from the raw signals. After exploring several neural network architectural setups, a proper convolutional neural network (CNN) was used for the initial diagnosis. Subsequently, using Wasserstein GAN with Gradient Penalty (WGAN-GP) and Variational Autoencoder (VAE), two different synthetic datasets were generated in order to augment the initial dataset and address the over-fitting issue. The augmented dataset using VAE achieved a 3.0\% improvement in accuracy reaching up to 99.0\% and yielded a lower loss value as well as a faster convergence. Finally, we addressed the lack of trust in black-box models using the Local Interpretable Model-agnostic Explanations (LIME) algorithm to determine the most important superpixels (frequencies) in the diagnosis process.

An Explainable Deep Learning-Based Method For Schizophrenia Diagnosis Using Generative Data-Augmentation

TL;DR

This work addresses the lack of objective diagnostic tools for schizophrenia by leveraging EEG-derived spectrograms and a CNN classifier trained with generative data augmentation. It compares VAEs and WGAN-GP for synthetic data, finding that a VAE-based augmentation (VAE-700) substantially improves accuracy to 99.0% on a 16-channel dataset, while the 19-channel dataset already achieves 99.6% accuracy without augmentation. The paper also employs LIME to provide explanations, revealing frequency-region patterns that differentiate schizophrenic and healthy EEG spectra, thereby enhancing trust in the model. Overall, the study demonstrates that targeted generative augmentation can generalize well on EEG-based schizophrenia diagnosis and that explainability methods help interpret neural decisions for clinical use.

Abstract

In this study, we leverage a deep learning-based method for the automatic diagnosis of schizophrenia using EEG brain recordings. This approach utilizes generative data augmentation, a powerful technique that enhances the accuracy of the diagnosis. To enable the utilization of time-frequency features, spectrograms were extracted from the raw signals. After exploring several neural network architectural setups, a proper convolutional neural network (CNN) was used for the initial diagnosis. Subsequently, using Wasserstein GAN with Gradient Penalty (WGAN-GP) and Variational Autoencoder (VAE), two different synthetic datasets were generated in order to augment the initial dataset and address the over-fitting issue. The augmented dataset using VAE achieved a 3.0\% improvement in accuracy reaching up to 99.0\% and yielded a lower loss value as well as a faster convergence. Finally, we addressed the lack of trust in black-box models using the Local Interpretable Model-agnostic Explanations (LIME) algorithm to determine the most important superpixels (frequencies) in the diagnosis process.
Paper Structure (21 sections, 4 equations, 14 figures, 9 tables, 1 algorithm)

This paper contains 21 sections, 4 equations, 14 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Our Contributions
  4. Preliminaries
  5. Data Augmentation Using Generative Models
  6. Wasserstein GAN with Gradient Penalty
  7. LIME Algorithm
  8. Methodology
  9. Datasets
  10. EEG Data Processing
  11. Classification Using Real Spectrograms
  12. Generating Synthetic Spectrograms
  13. Explanations by LIME
  14. Experiments and Results
  15. Initial Classification
  16. ...and 6 more sections

Figures (14)

  • Figure 1: A 16-channel EEG signal with 7680 samples of (a) healthy subject, and (b) schizophrenic patient
  • Figure 2: A 19-channel EEG signal with 185000 samples of (a) healthy subject, and (b) schizophrenic patient
  • Figure 3: The spectrograms of 5-second segments obtained from: (a) 16-channel dataset, and (b) 19-channel dataset.
  • Figure 4: An overview of the methodology used for diagnosis and explanation. The bottom branch represents the synthetic data generations
  • Figure 5: The proposed CNN architecture
  • ...and 9 more figures