Life Sciences & Biotechnology
Title : | Data-Driven Approaches for Enhanced Detection of Fetal Arrhythmia |
Area of research : | Life Sciences & Biotechnology |
Principal Investigator : | Dr. Muthukumar KA, University Of Petroleum And Energy studies (UPEs), Dehradun, Uttarakhand |
Timeline Start Year : | 2024 |
Timeline End Year : | 2027 |
Contact info : | muthukumar9890@gmail.com |
Equipments : | Worksation and Graphics card |
Details
Executive Summary : | The project aims to develop advanced techniques for fetal arrhythmia detection and fetal electrocardiogram (FECG) signal quality assessment. The health of the unborn child can be jeopardized by fetal arrhythmia, so early detection is essential. Traditional arrhythmia detection methods rely on accurate heartbeat detection, which can result in decreased accuracy. To solve this problem, we propose a deep learning-based strategy that makes use of FECG signals. The FECG signals are segmented and fed into a deep learning model, where they are classified as either normal or arrhythmic. A subject can be classified as healthy or arrhythmic after analyzing multiple segments. The estimated heartbeat interval is used to categorize the training data into normal, arrhythmic, and possibly mixed segments. The classification accuracy of the deep learning model is increased by training it with only the labelled normal and arrhythmic segments. This strategy lessens the reliance placed on the detection of an accurate heartbeat, which ultimately leads to a more reliable identification of fetal arrhythmia. Furthermore, the project addresses the issue of assessing signal quality in fetal ECG recordings. Earth Mover's Distance (EMD) can be used to compare the distribution of observed fetal ECG signals with a reference distribution in the context of fetal ECG signal quality assessment. High-quality fetal ECG signals or a model distribution of the ideal fetal ECG waveform can be used to build the reference distribution. Dissimilarity can be quantified by determining the expected difference (EMD) between the observed fetal ECG signal and the reference distribution. The dissimilarity between the observed signal and the reference distribution is quantified using this distance metric. Unsupervised learning techniques, such as clustering and self-organizing maps, can be applied to the EMD-based method to classify segments of fetal ECG signals with similar EMD values. This is useful for distinguishing between segments of high-, medium-, and low-quality signals. Using EMD for fetal ECG signal quality assessment eliminates the need for labelled training data, allowing for an objective evaluation of the similarity between observed signals and reference distributions. In order to boost the precision of downstream analysis tasks like fetal heart rate estimation and arrhythmia detection, this method can help identify low-quality segments, which can then be removed or flagged. The potential for this work to enhance the detection of fetal arrhythmia and the evaluation of signal quality in fetal ECG recordings is what makes it so important. Improved prenatal care and fetal health outcomes are possible with the proposed strategies. By utilizing deep learning techniques for arrhythmia detection and introducing an EMD approach for fetal ECG signal quality assessment, the project contributes to the field of research. |
Total Budget (INR): | 22,15,004 |
Organizations involved