Cross-wavelet assisted convolution neural network (AlexNet) approach for phonocardiogram signals classification

The exponential growth of a multitude of cardiovascular diseases, leading to life frightening conditions, makes fast and accurate computer-aided techniques that are relevant and important to identify these lethally medical anomalies. The authors aim to build up an accurate scheme that analyzes the Phonocardiogram (PCG) signal and find out whether the patient's heart works normally or required any special intervention for further diagnosis. This paper has a universal, object - to aid doctors and medical personnel, and an indispensable technique like a Cross-wavelet transform (XWT) assisted Convolution neural network (CNN) utilizing the AlexNet model to detect abnormal heart sounds which are the symbol of cardiovascular disease. A pre-trained AlexNet model has been used and fine-tuned to improve system performance. Convolution neural network (Alex Net architecture) utilizes the Cross-wavelet spectrum image as an input, to prevent and protect individuals from fatal medical conditions. The proposed method is applied both on the raw PCG data and on the PCG data after removing the noise. The results of the study show that our technique attained an accuracy of 98% and 97.89% when processed with the raw and the de-noised PCG dataset to distinguish abnormal heart sound from normal ones and outperformed all existing methods. The authors used the same dataset and evaluated three classic classifiers - LVQ, LS-SVM and PNN to compare the results with the proposed classifier. Also, the works which have been published by the previous researchers are sited and compared with our proposed approach.

Automated summary

Authors aim to analyze Phonocardiogram (PCG) signal using a Convolution neural network (CNN) with Cross-wavelet transform (XWT) to detect abnormal heart sounds, achieving an accuracy of 98% and outperforming existing methods. They used a pre-trained AlexNet model and evaluated classic classifiers to compare results.