Abstract
Multimodal machine learning is a recent yet fast-growing topic in brain disorders research. The aim is to capitalize on the complementary nature of different modalities to build better prediction models for psychiatric and neurologic disorders. In this chapter, we focus on three groups of methods for multimodal integration that differ with respect to the stage in which they are implemented along the machine learning pipeline: early, intermediate, or late. We first describe data fusion as an early data integration method. This is an emerging and promising approach that involves the integration of the multiple modalities at an early stage of the pipeline (e.g., feature extraction stage). We then discuss kernel-based methods and deep learning as intermediate data integration methods. Finally, we discuss ensemble methods as examples of late data integration methods. In the last section, we present some exemplar studies that have used multimodal machine learning to investigate brain disorders.
Original language | English |
---|---|
Title of host publication | Machine Learning |
Subtitle of host publication | Methods and Applications to Brain Disorders |
Publisher | Elsevier |
Pages | 283-305 |
Number of pages | 23 |
ISBN (Electronic) | 9780128157398 |
ISBN (Print) | 9780128157398 |
DOIs | |
Publication status | Published - 1 Jan 2019 |
Keywords
- Alzheimer’s disease
- Brain disorders
- Data fusion
- Ensemble
- Machine learning
- Multi-kernel learning
- Multimodal
- Neuroimaging
- Psychosis
- Sum of kernels