Welcome to the official repository for the Machine Learning in Neuroimaging course. This program is designed to bridge the gap between raw neurobiological data and predictive analytics, providing a rigorous mathematical and computational foundation for the next generation of neuroscientists.
This repository contains curated materials, code implementations, and theoretical frameworks for a 6-session intensive journey through the neuroimaging machine learning pipeline. We transition from high-dimensional imaging features to robust, interpretable, and generalizable predictive models.
📂 Session 01 | The Foundation
Neuroimaging ML Pipeline, Data Preprocessing, and EDA
From imaging-derived features to clean, analysis-ready machine learning datasets.
- Data cleaning strategies for voxel-wise and ROI-based metrics.
- Handling missing values and outlier detection in clinical cohorts.
- Exploratory Data Analysis (EDA) for neuroimaging distributions.
📂 Session 02 | The Theory
Machine Learning Foundations: Optimization, Generalization, and Regularization
Core mathematical and conceptual principles behind reliable predictive modeling.
- Loss functions and gradient-based optimization.
- The Bias-Variance tradeoff in brain-behavior mapping.
- Preventing overfitting via
$L_1$ (Lasso) and$L_2$ (Ridge) penalties.
📂 Session 03 | Filtering & Linear Models
Filter-Based Feature Selection with Linear and Kernel Supervised Models
- Statistical feature ranking (T-tests, ANOVA, Correlation).
- Implementation of Logistic Regression and SVMs for diagnostic classification.
- Applying K-Nearest Neighbors (KNN) on manifold structures.
📂 Session 04 | Non-Linear Complexity
Embedded Feature Selection with Tree-Based and Ensemble Models
- Decision Trees and the logic of recursive partitioning.
- Random Forests and Gradient Boosting for capturing non-linear brain patterns.
- Feature importance derived directly from tree-based architectures.
📂 Session 05 | Latent Spaces
Extraction-Based Feature Selection: Unsupervised Learning & Dimensionality Reduction
- PCA & ICA: Decomposing neural signals into independent components.
- Manifold Learning: Visualizing high-dimensional neuroimaging data.
- Clustering: Identifying data-driven sub-types in heterogeneous disorders.
📂 Session 06 | Advanced Refinement
Wrapper Feature Selection & Robust Model Evaluation
- Recursive Feature Elimination (RFE) and Sequential Feature Selection.
- Hyperparameter tuning via Grid/Random Search.
- Nested Cross-Validation for unbiased performance estimation in neuroimaging.
- Languages: Python (Primary)
- Libraries:
scikit-learn,pandas,numpy,seaborn - Recommended Environment: Jupyter Lab or Google Colab
This repository serves as a live document for the IDS Mentorship Program. Participants are encouraged to:
- Clone the repository.
- Complete the weekly assignments in the
sessions/directory. - Submit Pull Requests for peer-review during lab hours.
Created with passion for the intersection of Brain Science and Artificial Intelligence.