Student Projects

available

We welcome spontaneous propositions of student projects related to the Biomedical Signal Processing course.
 
Master project – Frequency estimators for biomedical signals:
Most of the biomedical signals are composed of time-varying oscillations. Precious information about subject’s health is usually contained in these biological rhythms. Although a large number of approaches operating either in the time-domain or in the frequency-domain have already been proposed, the real-time estimation of the time-varying frequency of such signals remains a challenge.  
The aim of this project is to investigate and implement some frequency estimation methods and evaluate their performance. More specifically, the implemented techniques will be used to estimate heart rate from video sequences, using a very recent technology called imaging photoplethysmography, which makes possible the remote sensing of blood volume changes occurring at each heartbeat from facial images of the subjects.
 Requirements: Signal processing, Matlab

 

Extraction and analysis of baroreflex activity from RR-interval and blood pressure recordings 
Baroreflex activity, which is expressed as a time-varying oscillation in the LF band (0.04 – 0.15 Hz) of RR-intervals and blood pressure, is central to cardiovascular regulation. The goal of this project is, using advanced adaptive frequency tracking algorithms, to extract this activity from RR-intervals and blood pressure sequences recorded in healthy and sick subjects to assess the differences in cardiovascular regulation induced by various pathologies such as diabetes.
Requirements: signal processing, good Matlab skills. 
 
Semester Project: Pattern Classification with Imperfect Labels

Classification techniques have been widely used in various pattern recognition and machine learning systems. However, fundamental issue with this approach is label imperfections in training data, since the line of demarcation between classes is determined based on field expert experience and maintenance capability. In other words, If the labels are not 100% correct, training will be done using imprecise knowledge and the classification accuracy decreases.  To address this issue we propose a noisy-label scenario in which the labeling noise is triggered by users reporting their emotion on a scale from 1 to 10 without being 100% sure. This semester projects studies and implements various techniques based on possibility theory and Dempster–Shafer theory of evidence to tackle the classification problem with imperfect labels. The proposed technique gives encouraging results on two industrial fault-prediction data sets. There is no need to acquire data for this project and we will use a dataset, which is available.

Tasks:

  • Reviewing the literature on classification of human emotions
  • Studying state of the art for handling imperfect and noisy labels
  • Studying and understanding possibility theory and Dempster–Shafer theory of evidence
  • Implementing the selected methodology for emotion classification using imperfect labels
  • Studying an available dataset and state of the art methodology to analyze it.

 

Requirements: signal processing, good Matlab skills. 

Contact: Ashkan Yazdani, Jean-Marc Vesin

 

Semester Project: Identification of Information Maximization Time in Emotion Detection using Brain and Physiological Signals

Assessment of users’ emotional state while watching or listening to multimedia content is becoming more and more important in many applications.  An approach for emotion assessment is to use the dimensional models of emotion such as those based on the arousal and valence dimensions. The estimated values of arousal and valence can be used to infer different emotional categories. This project involves research on the classification of positive/negative valence, high/low arousal, and like/dislike, induced in users while viewing different music video clips, based on the analysis of the EEG and the peripheral physiological signals. High speed physiological signal processing is of significant importance in human computer interface (HCI) systems that are developed for real-time communication. The recorded signals can be processed, before feature extraction, to evaluate and identify the time point when the signals convey maximum emotional information, which can be then used for detection or classification. According to the information theory, instantaneous entropy can be a powerful tool for this evaluation. The aim of this project is to detect the information maximization time in EEG and peripheral physiological signals during different emotional states and to use the corresponding segments to evaluate the overall performance of the emotion detection system and compare the result with previous research. There is no need to acquire data for this project and we will use a dataset, which is available.

Tasks:

  • Reviewing the literature about emotion analysis using physiological signals
  • Studying state of the art for measuring real time information maximization time 
  • Implementing the selected methodology for detecting information maximization in various emotional states and to classify them.
  • Studying an available dataset and state of the art methodology to analyze it. 

Requirements: signal processing, good Matlab skills. 

Contact: Ashkan YazdaniJean-Marc Vesin

 

previous-student-projects

Semester projects

  • De-noising of the electrocardiogram, spring 2014, by David Blanchet and Bertrand Tricherini
  • Estimation of the instantaneous frequencies of cardio-respiratory oscillations, automn 2013, by Laurent D’Andres
  • Waveform characterization of atrial fibrillation, spring 2010
  • Optimization of dominant frequency trajectory estimation in time-frequency planes,spring 2010
  • Development of complexity measures with applications to biomedical signals, spring 2010
  • Poursuite de fréquences et segmentation de signaux de motilité gastro-intestinale, spring 2008, by Anouk André        
  • Fibrillation auriculaire – Poursuite en fréquence et classification, spring 2008, by Frank de Morsier
  • Analyse de la dynamique des intervalle RR durant la fibrillation auriculaire, spring 2008 by Ricardo Esquil   

 

Master projects

  • Development of complexity measures with applications to biomedical signals, spring 2010
  • Adaptive tracking of harmonic frequencies, spring 2010
  • Physiological wireless measurements, spring 2008, by Olivier Grossenbacher