Inception-Based Network and Multi-Spectrogram Ensemble Applied To Predict Respiratory Anomalies and Lung Diseases

Pham, Lam; Phan, Huy; Strobel, Alexander; King, Ross; Mertins, Alfred; McLoughlin, Ian

doi:10.1109/embc46164.2021.9629857

Cited by 16 publications

(8 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As the potential results were achieved from our previous work [7], we further evaluate different inception based network architectures in this paper. In particular, two high-level architectures with single or double inception layers as shown in Table I are used in this paper.…”

Section: B the Back-end Deep Learning Network For Classification (I) ...mentioning

confidence: 99%

“…As proposed deep learning frameworks are shown in Figure 1, we firstly duplicate the short-time cycles or cut off the long-time cycles to make all respiratory cycles equal to 10 seconds. For the first two deep learning frameworks (I) and (II), we extract Wavelet-based spectrograms which proves effective in our previous work [7]. As we reuse the setting from [7], we then generate Wavelet spectrograms of 124×154 from 10-second respiratory cycles.…”

Section: A the Front-end Spectrogram Feature Extractionmentioning

confidence: 99%

“…For the first two deep learning frameworks (I) and (II), we extract Wavelet-based spectrograms which proves effective in our previous work [7]. As we reuse the setting from [7], we then generate Wavelet spectrograms of 124×154 from 10-second respiratory cycles. Meanwhile, we extract log-Mel spectrogram in the deep learning framework (III) as we re-use pre-trained models from [14] receiving log-Mel spectrogram input.…”

Section: A the Front-end Spectrogram Feature Extractionmentioning

confidence: 99%

“…Meanwhile, deep learning based systems make use of spectrogram where both temporal and spectral features are well represented. These spectrograms are then explored by a wide range of convolutional deep neural networks (CNNs) [4], [5], [6], [7] or recurrent neural networks (RNNs) [8], [9]. Compare between two approaches, deep learning based systems are more effective and show potential results mentioned in [4], [6], [7].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Robust Deep Learning Framework For Predicting Respiratory Anomalies and Diseases

Pham

McLoughlin

Phan

et al. 2020

2020 42nd Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

Self Cite

View full text Add to dashboard Cite

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.

show abstract

Section: B the Back-end Deep Learning Network For Classification (I) ...mentioning

confidence: 99%

Section: A the Front-end Spectrogram Feature Extractionmentioning

confidence: 99%

Section: A the Front-end Spectrogram Feature Extractionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Robust Deep Learning Framework For Predicting Respiratory Anomalies and Diseases

Pham

McLoughlin

Phan

et al. 2020

2020 42nd Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Meanwhile, deep learning based systems make use of raw inputs such as waveforms or spectrograms, with a trained feature extractor. Spectrograms, in which both temporal and spectral feature elements are well represented, have been explored by a wide range of deep and convolutional neural networks (CNNs) [4], [5], [6], [7] and recurrent neural networks (RNNs) [8]. Comparing between machine learning approaches with hand-crafted features, and deep learning systems with trained feature extractors, the latter are widely reported as being more effective for respiratory classification tasks [4], [6], [7].…”

Section: Introductionmentioning

confidence: 99%

An Ensemble of Deep Learning Frameworks for Predicting Respiratory Anomalies

Pham

Ngo

Tran

et al. 2022

2022 44th Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

Self Cite

View full text Add to dashboard Cite

This paper evaluates a range of deep learning frameworks for detecting respiratory anomalies from input audio. Audio recordings of respiratory cycles collected from patients are transformed into time-frequency spectrograms to serve as front-end two-dimensional features. Cropped spectrogram segments are then used to train a range of back-end deep learning networks to classify respiratory cycles into predefined medically-relevant categories. A set of those trained high-performance deep learning frameworks are then fused to obtain the best score. Our experiments on the ICBHI benchmark dataset achieve the highest ICBHI score to date of 57.3%. This is derived from a late fusion of inception based and transfer learning based deep learning frameworks, easily outperforming other state-of-the-art systems.

show abstract