Tutorial: Deep Learning Workflows for Biomedical Signal Data – a Practical Example
The use of AI techniques on signals is growing in popularity across different industries for a variety of applications, including the medical and healthcare areas, such as digital health, physiological signal analysis, and patient monitoring applications.
In this interactive technical session, we'll use real-world ECG datasets to demonstrate deep learning approaches in MATLAB to show sequence-to-sequence classification frameworks for 1-D signals. Some of the tasks we'll explore in developing advanced predictive models include:
- semi-automated data labeling strategies
- (automated) feature extraction techniques
- transformations to the time-frequency domain
- comparing model topology choices and model parameter tuning.
We will follow each section with an open discussion and exchange of experiences.
Target audience and expected prerequisite knowledge
The target audience of this tutorial are practitioners of artificial intelligence interested in workflow discussions based on hands-on examples. The presented workflows generalize beyond ECG datasets to other types of signals and time-series data. There are no prerequisites other than general knowledge of deep learning methodologies.
In this session we will explain basic theoretical concepts and then show individual workflow steps in MATLAB based on executable code in a notebook format, and various tools and apps MATLAB offers in this regard.
We will provide an overview of some of the most frequently used techniques including transformations to the time-frequency domain, such as Wavelet or Hilbert-Huang transformations. In terms of model architectures, we will consider recurrent neural networks (specifically LSTMs).
We will illustrate the workflows on an ECG (electrocardiogram) dataset using sequence-to-sequence classification (ECG signal segmentation). The relative length and morphologies of the major characteristics of ECG signals are of paramount importance to the diagnosis and treatment of various cardiac diseases.
Christoph Stockhammer holds a M.Sc. degree in Mathematics from the Technical University Munich with an emphasis on optimization. He joined The MathWorks in 2012 and works as an application engineer. His focus areas include Mathematics and data analytics, Machine & Deep Learning as well as the integration of MATLAB software components in other programming languages and environments. email@example.com
Dr Mihaela Jarema is part of the Academia Group at MathWorks in Munich/Germany. She partners with research institutes in Germany to accelerate their discovery and learning. Mihaela holds a PhD degree in computer science from Technische Universität München. During her PhD, she has used AI techniques with MATLAB to model ensemble data, evaluate, and visualize the associated variability. firstname.lastname@example.org