Abstract

Machine learning (ML) has been successfully applied to analyze connectomic data and construct predictive models. These models can help identify disease biomarkers, track disease progression, and quantify the risk of developing a condition at an earlier stage than conventional diagnostic techniques. In this chapter, we discuss how ML methods can be used to address a variety of tasks in the context of connectomic analysis. We begin by defining different types of ML paradigms and present a standard ML workflow for connectomic data. We, then, explore different approaches to obtain input representations required to feed the connectomic graphs into an ML algorithm and discuss relevant methods, such as graph kernels, linear classifiers and deep neural networks, to model a prediction problem. The chapter concludes by discussing important metrics used to summarize the performance of ML models, and provides a list of good practices that should be followed to ensure performance estimates are unbiased and generalizable.

Original languageEnglish
Title of host publicationConnectome Analysis
Subtitle of host publicationCharacterization, Methods, and Analysis
PublisherElsevier
Pages267-287
Number of pages21
ISBN (Electronic)9780323852807
ISBN (Print)9780323852814
DOIs
StatePublished - Jan 1 2023

Keywords

  • algorithm
  • connectomics
  • kernel
  • Machine learning
  • model fitting
  • neural networks

Fingerprint

Dive into the research topics of 'Machine learning in connectomics: from representation learning to model fitting'. Together they form a unique fingerprint.

Cite this