An Ensemble of Deep Learning Frameworks Applied For Predicting Respiratory Anomalies

TL;DR

This study evaluates and combines multiple deep learning models to improve the detection of respiratory anomalies from audio recordings, achieving state-of-the-art results on a benchmark dataset.

Contribution

It introduces a fusion approach of deep learning frameworks applied to spectrograms for respiratory anomaly detection, outperforming existing methods.

Findings

Achieved highest ICBHI score of 57.3 with late fusion.

Spectrogram-based features effectively represent respiratory cycles.

Fusion of inception and transfer learning models improves accuracy.

Abstract

In this paper, we evaluate various deep learning frameworks for detecting respiratory anomalies from input audio recordings. To this end, we firstly transform audio respiratory cycles collected from patients into spectrograms where both temporal and spectral features are presented, referred to as the front-end feature extraction. We then feed the spectrograms into back-end deep learning networks for classifying these respiratory cycles into certain categories. Finally, results from high-performed deep learning frameworks are fused to obtain the best score. Our experiments on ICBHI benchmark dataset achieve the highest ICBHI score of 57.3 from a late fusion of inception based and transfer learning based deep learning frameworks, which outperforms the state-of-the-art systems.