January 31, 2020IEEE Journal of Biomedical and Health InformaticsOpen Access

Towards Domain Invariant Heart Sound Abnormality Detection Using Learnable Filterbanks

View Full Paper

Ask AI

Bookmark

Why the study?

Machine learning systems for heart sound abnormality detection lack robustness due to domain variability from stethoscopes, environments, and data collection protocols.

Population

Publicly available multi-domain datasets

Comparison

CNN with learnable filterbank front-end vs top-scoring systems from the literature

Design

Algorithm development and validation study

Authors

AHAhmed Imtiaz HumayunBangladesh University of Engineering and Technology SGShabnam GhaffarzadeganAmirkabir University of Technology MAMd. Istiaq AnsariUniversity of Central Florida

Discussion

Member takes

Implication

Domain shifts impair heart sound classifiers; learnable filterbanks support cross-domain robustness but require prospective trials before clinical use.

Structured PICO

Population

1087 patients contributing 2599 normal and 1199 abnormal phonocardiogram (PCG) recordings from the 2016 PhysioNet/CinC Challenge Database and 2018 INTERSPEECH ComParE Dataset.

Intervention

Convolutional Neural Network (CNN) with a learnable filterbank front-end consisting of time-convolutional (tConv) units

Comparator

State-of-the-art baseline machine learning methods (e.g., bagged Support Vector Classification ensembles, static front-end 1D-CNN)

Outcome

Heart sound abnormality detection performance (measured by sensitivity, specificity, F-1 score, and MAcc)

A novel learnable filterbank CNN architecture improves the robustness of automated heart sound abnormality detection across different stethoscopes and sensors.

Cite This Study

Humayun et al. (2020) studied this question.

synapsesocial.com/papers/6a7668214fe2e6ffd8ec004f https://doi.org/10.1109/jbhi.2020.2970252

View Full Paper

Ask AI

Bookmark

Implication

Domain shifts impair heart sound classifiers; learnable filterbanks support cross-domain robustness but require prospective trials before clinical use.

Towards Domain Invariant Heart Sound Abnormality Detection Using Learnable Filterbanks

Discussion

Member takes

Implication

Structured PICO

Cite This Study

Towards Domain Invariant Heart Sound Abnormality Detection Using Learnable Filterbanks

Discussion

Member takes

Implication

Structured PICO

Cite This Study

Also Consider

Also Consider