Docstoc

To Whom It May Concern To Whom It May Concern This letter is to

Document Sample
To Whom It May Concern To Whom It May Concern This letter is to Powered By Docstoc
					                                      To Whom It May Concern
This letter is to recommend the thesis “A Decision Support System for Stress Diagnosis using ECG Sensor”, by
Mohd Siblee Islam, for the SAIS award for best Masters thesis in the area of AI. The main objective of the thesis
is to construct a Case-Based Reasoning system based on heart rate variability features to diagnose stress levels.
The student successfully defended his thesis in November 2010.

Despite a complex and difficult problem in the medical domain the thesis has been carried out systematically and
successfully to solve the problem by investing several AI techniques (like experience-based outlier removal,
CBR and fuzzy logic) together with signal processing methods (i.e. FFT and DFT).

The thesis shows how AI techniques can be applied to develop medical systems that satisfy real-world problems
and the thesis will be a valuable contribution and part of an ongoing EU FP7 research project, which the
Intelligent Systems Group at Mälardalen University participates in.

The contribution of this thesis work is that:
    •    Multi-domain features are used in a creative and novel approach to diagnose stress from sensor readings
         i.e. time domain and frequency domain features are extracted using Fast Fourier Transformation and
         time-frequency domain features are extracted using Discrete Wavelet Transformation.
    •    In signal preprocessing, a novel experience based outlier removing algorithm has been applied to detect
         outliers and replace them. It is a completely new approach instead of manual visual inspection and
         interpolation of data based on expert’s observation and experience.
    •    A Case-Based Reasoning approach has been applied together with a Fuzzy logic approach and several
         signal processing techniques to design, develop and implement a case-based system to diagnosis
         individual stress. The system was evaluated with the close collaboration of an expert in the domain and
         the weights of the features are tuned to get a better performance.
    •    A focused literature study has been done to find out possible domains to extract features from an Inter-
         beat-interval signal and for feature selection i.e. to distinguish features according to their importance.
    •    Evaluation shows that the accuracy achieved is high (86%) where the sensitivity is relatively low
         (57%). However, the thesis provided sufficient proof that it is due to a small number of available stress
         cases (i.e. 7 out of 22 cases). Moreover, the possible ways to improve the sensitivity has been
         investigated and clearly explained and presented in the report.
    •    A java based software tool has been developed as a research prototype to validate the functionalities of
         the proposed system i.e. collect ECG signals through ECG sensors, remove outliers, calculate and
         extract features for all three domains. In addition, a web-based framework has also been develop to
         facilitate the CBR approach in diagnosing stress i.e. multi-domain case retrieval, case and weight of the
         features’ maintenance.


For these reasons above, we recommend the thesis to be a candidate for the SAIS thesis award.




Supervisors & Examiner
Shahina Begum, Mobyen Ahmed and Peter Funk
Shahina.begum@mdh.se
School of innovation, design and engineering
Intelligent Systems Group
Mälardalen University, Västerås, Sweden

				
DOCUMENT INFO
Shared By:
Tags:
Stats:
views:47
posted:11/2/2011
language:English
pages:1