Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating Neural Networks and Dictionary Learning for Multidimensional Clinical Characterizations from Functional Connectomics Data

Niharika Shimona D'Souza, Mary Beth Nebel, Nicholas Wymbs, Stewart Mostofsky, Archana Venkataraman

TL;DR

A unified optimization framework that combines neural networks with dictionary learning to model complex interactions between resting state functional MRI and behavioral data outperforms state-of-the-art methods in a ten-fold cross validated setting to predict three different measures of clinical severity.

Abstract

We propose a unified optimization framework that combines neural networks with dictionary learning to model complex interactions between resting state functional MRI and behavioral data. The dictionary learning objective decomposes patient correlation matrices into a collection of shared basis networks and subject-specific loadings. These subject-specific features are simultaneously input into a neural network that predicts multidimensional clinical information. Our novel optimization framework combines the gradient information from the neural network with that of a conventional matrix factorization objective. This procedure collectively estimates the basis networks, subject loadings, and neural network weights most informative of clinical severity. We evaluate our combined model on a multi-score prediction task using 52 patients diagnosed with Autism Spectrum Disorder (ASD). Our integrated framework outperforms state-of-the-art methods in a ten-fold cross validated setting to predict three different measures of clinical severity.

Integrating Neural Networks and Dictionary Learning for Multidimensional Clinical Characterizations from Functional Connectomics Data

TL;DR

A unified optimization framework that combines neural networks with dictionary learning to model complex interactions between resting state functional MRI and behavioral data outperforms state-of-the-art methods in a ten-fold cross validated setting to predict three different measures of clinical severity.

Abstract

We propose a unified optimization framework that combines neural networks with dictionary learning to model complex interactions between resting state functional MRI and behavioral data. The dictionary learning objective decomposes patient correlation matrices into a collection of shared basis networks and subject-specific loadings. These subject-specific features are simultaneously input into a neural network that predicts multidimensional clinical information. Our novel optimization framework combines the gradient information from the neural network with that of a conventional matrix factorization objective. This procedure collectively estimates the basis networks, subject loadings, and neural network weights most informative of clinical severity. We evaluate our combined model on a multi-score prediction task using 52 patients diagnosed with Autism Spectrum Disorder (ASD). Our integrated framework outperforms state-of-the-art methods in a ten-fold cross validated setting to predict three different measures of clinical severity.

Paper Structure

This paper contains 15 sections, 6 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Multidimensional Clinical Characterization from Functional Connectomics
  3. Joint Optimization Strategy:
  4. Proximal Gradient Descent on $\mathbf{X}$.
  5. Updating the Neural Network Weights $\mathbf{\Theta}$.
  6. L-BFGS update for $\{\mathbf{c}_{n}\}$.
  7. Augmented Lagrangian Update for the Constraint Variables.
  8. Prediction on Unseen Data.
  9. Baseline Comparisons
  10. Implementation Details.
  11. Experimental Evaluation and Results:
  12. Data and Preprocessing.
  13. Performance Characterization.
  14. Conclusion
  15. Acknowledgements.

Figures (3)

  • Figure 1: A unified framework for integrating neural networks and dictionary learning. Blue Box: Dictionary Learning from correlation matrices Gray Box: Neural Network architecture for multidimensional score prediction
  • Figure 2: Multi-Score Prediction performance for Top: ADOS Middle: SRS Bottom: Praxis by Red Box: Our Framework. Green Box: Generative-Discriminative Framework from d2018generative. Blue Box: BrainNet CNN from kawahara2017brainnetcnn
  • Figure 3: Eight subnetworks identified by our model from multi-score prediction. The blue and green regions are anticorrelated with the red and orange regions.