Research

Automated systems for detecting neonatal brain injuries (contextual classification)

Brain injury at the time of birth may cause neural dysfunctions or death in severe cases. The problem with detecting the brain injuries in neonates is that they don’t show any clear clinical signs of the brain damage. Recording the electrical activity of brain called Electroencephalogram (EEG)/(brain signals) is considered the gold standard for the detecting any brain injuries and monitoring any progression or improvement. However, expertise to interpret the neonatal EEG is scarce in busy Neonatal Intensive Care Units (NICU). Thus an automated system to detect brain injuries in neonates could help clinical staff in diagnosis and to suggest an early treatment.

A lot of work has been done in past to develop such automated systems for neonates but most of them used short segments of these brain signals to detect any brain malfunctioning events. However it is well known that there is a lot of information in brain signals if looked at longer time scale. Therefore, in the current work a novel automated system is being explored that can look at this subtle but significantly important contextual information. The results show promising performance improvements in the detection of brain injuries.

More specifically, different dynamic kernels for Support vector machine are explored. Additionally, other classification approaches e.g Gaussian mixture model with universal background models are also investigated.

Currently I am working towards a multi-modal and multi-stream system where data from different physiological signals will be used together to get an overall score of the neonatal brain health. (Publications, Doctoral Thesis)

An example of evolving seizure (brain injury event)

Transforming a classifier that can only classify short segments (fixed length feature vector) to a classifier that can classify sequences of such short segments (variable length).

Wireless textile sweat detector (zigbee based technology)

(Master's Thesis)

Hemodialysis is used to clean blood and remove extra fluid from the body of patient suffering renal failure. During hemodialysis, dehydration could occur, if fluid from the body is removed aggressively. This leads to unconsciousness of patient. It is observed that just before this condition occurs, sweating starts on palms and forehead of patient. So sweat detection could serve an early warning system to this unusual condition.The purpose of this project was to develop a sweat detector for hemodialysis patient, which can wirelessly transmit the sweat sensor information to caregiver.

A new textile sensor prepared with conductive fibre weaved in different patterns, was designed to detect sweating. This sensor was connected with the PICDEM Z motherboard, which digitizes the sensor information, and sends it to data communication.

The results of the developed sensor showed good resistance drop characteristics against sweat. So this sensor could be used for detecting sweat of a hemodialysis patient. Software and hardware designed for this project is proved to have a reliable communication and processing capabilities during its test runs.

The code or stack for the implementation of zigbee protocol can be found here (Software, Documentation)

Textile sweat sensor

ZigBee wireless development Board

Feature extraction and classification of EBI measurements

(Term Project) This study analysed different features extracted from the focal measurement of EBI on the ankle from 3 different patients (Healthy, Mild, Svere). Feature analysis is done by using Cole model parameters, the impedance Indexes and the spectral features using MATLAB environment.

In this project we developed a k-NN classifier, whose implementation concept is also discussed in the (Report).

Min Max Plots Cole Plot

Heart rate analysis And visualization

(Term Project) The main purpose of this project was to present a practical approach for implementing heart rate analysis and visualization techniques in MATLAB, where the heart rate features are extracted from ECG signals using various signal processing methods. To better understand the system, intermediate goals have been included to describe the theory behind common heart rate analysis-, detection-, feature extraction-, and visualization techniques, along with a brief description of the physiological source of the signals.

A complete GUI was developed with the option of loading the files and visualising the R peaks as well as the poincare plot and heart rate interval plot. (ECG tutorial, Report)

The code R peaks detection and HRV visualization can be requested !

Modeling and simulation of optical mm-wave signal generation by single-side-band intensity-modulation for use in Radio Over Fiber (ROF) Links.

Fiber based wireless access systems support high-speed in real-time. By combining the optical and wireless networks we can have a network that has capacity of optical fiber and flexibility of wireless network. For example, radio-over-fiber (ROF) link can support wireless LAN and cellular radio simultaneously.

In this study, we simulated the transmission performance of the optical mm-wave generation by using an external modulator based on single sideband (SSB) intensity modulation.

Radio over fiber network concept

Page updated

Google Sites

Report abuse