Skip to content

Minicourse EDTH-JA19

AI-driven methods for the interpretable analysis of fNIRS signals in Python

Javier Andreu-Perez

with Javier Fumanal-Idocin, Jessica Caterson, Mohammadreza Jamalifard

About this course

In this mini-course, we will introduce a toolbox designed to utilise the latest AI-powered data analytics methods for multivariate pattern analysis of fNIRS data. Unlike other approaches, this mini-course focuses on exFuzzy, a valuable Python toolbox that generates interpretable logic statements directly from fNIRS data, streamlining the process from data collection to publication. The use of this toolbox is not exclusive, it can be integrated with the outputs other fNIRS toolboxes, such as Homer, AnalyzIR, SPM-fnirs or libraries/toolbox from other manufacturers. The toolbox can be used from data from other analytic toolboxes and can be exported in a tabular format for use. That can be read in Python. exFuzzy works similarly to sk-learn but provides extended features that do not exist in this library.

This toolbox offers an introduction to AI-powered multivariate analytics for fNIRS using logic-based artificial intelligence methods. These methods offer more explicit information compared to standard machine learning approaches, providing a framework for rapid data analysis using AI without the need to unravel complex models. This approach integrates well with classical fNIRS data analysis and standard data interrogation techniques.

Learning Outcomes

  • Learn the foundations of fuzzy logic
  • Dive into fNIRS data analysis with Python
  • Perform multivariate-pattern analysis

Course Plan

Level: Introductory

Pre-Requisites: Course requires software installation (click here for more information); basic knowledge of Python is desirable but not required (basics will be covered)
Course Duration: 2 hours

Delivery Plan

1st hour: An introductory exploration into the realm of symbolic artificial intelligence, covering model construction and interpretation. This session delves into the motivation behind this methodology and its current applications.
2nd hour: Hands-on experiment using fNIRS data, where the participants would be able to parametrise the models, work on the toy example, as well as discuss their own data analytics plan.

Why enrol on this course?

This minicourse aims to introduce the fNIRS community to the foundations of ex-fuzzy through hands-on exercises. It is recommended for those interested in using Machine Learning in their fNIRS research. Traditional Machine Learning and Deep Learning methods in neuroscience often face challenges for fNIRS analysis, as these models were designed for predictive outputs rather than analytical insights. Many fNIRS research applications, including those in neurodevelopmental cognitive neuroscience, healthcare, neuroergonomics and HCI, require accountable analysis, which current models do not provide without post hoc interpretation—a process that often results in coarse estimations. Interpretable symbolic models offer a new paradigm of analytics, enabling explicit traceability from data to outputs, incorporating underlying functional connectomes or logical statements that drive predictions.

Participants will also have the opportunity to discuss their own data with the teaching staff, provided the data is formatted for use in Python, in vectorial format.