Deep Learning to Predict Respiratory Flow from Ribcage Movements
Software in wearable technology relies on nonlinear transformations that predict state variables of the human body from sensor data. Analytical approaches based on signal processing and physics have reigned supreme for many decades for such prediction. However, the advent of deep learning of neural network architectures throws down the gauntlet to traditionalists and claims to automate discovery of complex nonlinear transformations purely from time-ordered input and output data while promising a steady improvement in accuracy with more data. This project will validate this claim by developing a deep learning architecture based on long short term memory networks to predict respiratory flow from the mouth/nose using measurements of ribcage movement. The student will use 240,000 training sequences from five subjects (all male, aged 26$\pm$1 years), performing submaximal and incremental physical tests on a cycle ergometer for deep learning and 14,000 test sequences to validate its performance. The student's goal will be to achieve a high accuracy in predicting respiratory flow.
The goal of the thesis is to develop and validate a deep learning model to predict respiratory flow from ribcage movements.
- Introduction to deep learning and IoT
- Product development for wearables
- Bachelors in computer science, sports science, or electrical engineering
- Good knowledge in Python programming and data analysis
- Sagar Sen
- Pierre Bernabe