Use this URL to cite or link to this record in EThOS:
Title: Intelligent ECG processing and abnormality detection using adaptive ensemble models
Author: Pandit, Diptandshu
ISNI:       0000 0004 7430 0639
Awarding Body: Northumbria University
Current Institution: Northumbria University
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This thesis explores the automated Electrocardiogram (ECG) signal analysis and the feasibility of using a set of computationally inexpensive algorithms to process raw ECG signals for abnormality detection. The work is divided into three main stages which serve towards the main aim of this research, i.e. the abnormality detection from single channel raw ECG signals. In the first stage, a lightweight baseline correction algorithm is proposed along with a modified moving window average method for real-time noise reduction. Additionally, for further offline analysis, a wavelet transform and adaptive thresholding based method is proposed for noise reduction to improve signal-to-noise ratio. In the second stage, a sliding window based lightweight algorithm is proposed for real-time heartbeat detection on the raw ECG signals. It includes max-min curve and dynamic (adaptive) threshold generation, and error correction. The thresholds are adapted automatically. Moreover, a sliding window based search strategy is also proposed for real-time feature extraction. Subsequently, a hybrid classifier is proposed, which embeds multiple ensemble methods, for abnormality classification in the final stage. It works as a meta classifier which generates multiple instances of base models to improve the overall classification accuracy. The proposed hybrid classifier is superior in performance, however, it is dedicated to offline processing owing to high computational complexity. Especially, the proposed hybrid classifier is also further extended to conduct novel class detection (i.e. unknown newly appeared abnormality types). A modified firefly algorithm is also proposed for parameter optimization to further improve the performance for novel class detection. The overall proposed system is evaluated using benchmark ECG databases to prove its efficiency. To illustrate the advantage of each key component, the proposed feature extraction, classification and optimization algorithms are compared with diverse state-of-the-art techniques. The empirical results indicate that the proposed algorithms show great superiority over existing methods.
Supervisor: Zhang, Li Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: G900 Others in Mathematical and Computing Sciences