5. Decoding and MVPA: predicting from brain images

Decoding consists in predicting external variables such as behavioral or phenotypic variables from brain image. It can be useful for diagnostic of prognosis, or to probe the information content of brain activity images.

These are Supervised learning tasks, focused on predicting an output value.

• 5.1. An introduction to decoding
  • 5.1.1. Loading and preparing the data
    • 5.1.1.1. The Haxby 2001 experiment
    • 5.1.1.2. Loading the data into nilearn
  • 5.1.2. Performing a simple decoding analysis
    • 5.1.2.1. A few definitions
    • 5.1.2.2. A first estimator
    • 5.1.2.3. Decoding made easy
    • 5.1.2.4. Measuring prediction performance
      • 5.1.2.4.1. Cross-validation
      • 5.1.2.4.2. Choosing a good cross-validation strategy
      • 5.1.2.4.3. Choice of the prediction accuracy measure
      • 5.1.2.4.4. Prediction accuracy at chance using simple strategies
    • 5.1.2.5. Visualizing the decoder’s weights
  • 5.1.3. Decoding without a mask: Anova-SVM
    • 5.1.3.1. Dimension reduction with feature selection
    • 5.1.3.2. Visualizing the results
    • 5.1.3.3. References
• 5.2. Choosing the right predictive model for neuroimaging
  • 5.2.1. Predictions: regression, classification and multi-class
    • 5.2.1.1. Regression
    • 5.2.1.2. Classification: two classes or multi-class
  • 5.2.2. Different linear models
  • 5.2.3. Setting estimator parameters
  • 5.2.4. Bagging several models
  • 5.2.5. References
• 5.3. FREM: fast ensembling of regularized models for robust decoding
  • 5.3.1. FREM pipeline
  • 5.3.2. Empirical comparisons
    • 5.3.2.1. Decoding performance increase on Haxby dataset
    • 5.3.2.2. Spatial regularization of decoding maps on mixed gambles study
  • 5.3.3. References
• 5.4. SpaceNet: decoding with spatial structure for better maps
  • 5.4.1. The SpaceNet decoder
  • 5.4.2. Related example
  • 5.4.3. References
• 5.5. Searchlight : finding voxels containing information
  • 5.5.1. Principle of the Searchlight
  • 5.5.2. Preparing the data
    • 5.5.2.1. Masking
  • 5.5.3. Setting up the searchlight
    • 5.5.3.1. Classifier
    • 5.5.3.2. Score function
    • 5.5.3.3. Cross validation
    • 5.5.3.4. Sphere radius
  • 5.5.4. Visualization
    • 5.5.4.1. Searchlight
    • 5.5.4.2. Comparing to massively univariate analysis: F_score or SPM
  • 5.5.5. References
• 5.6. Running scikit-learn functions for more control on the analysis
  • 5.6.1. Performing decoding with scikit-learn
    • 5.6.1.1. Using scikit-learn estimators
    • 5.6.1.2. Cross-validation with scikit-learn
    • 5.6.1.3. Measuring the chance level
  • 5.6.2. Going further with scikit-learn
    • 5.6.2.1. Decoding without a mask: Anova-SVM using scikit-learn
    • 5.6.2.2. Using any other model in the pipeline
  • 5.6.3. Setting estimator parameters