Open Access

ARTICLE

Recognition System for Diagnosing Pneumonia and Bronchitis Using Children’s Breathing Sounds Based on Transfer Learning

by Jianying Shi¹, Shengchao Chen¹, Benguo Yu², Yi Ren^3,*, Guanjun Wang^1,4,*, Chenyang Xue⁵

1 School of Information and Communication Engineering, Hainan University, Haikou, 570228, China
2 School of Biomedical Information and Engineering, Hainan Medical College, Haikou, 571199, China
3 Department of Pediatrics, Haikou Hospital of the Maternal and Child Health, Haikou, 570203, China
4 Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, Wuhan, 430074, China
5 School of Instrument and Electronics, North University of China, Taiyuan, 030051, China

* Corresponding Authors: Yi Ren. Email: ; Guanjun Wang. Email:

(This article belongs to the Special Issue: Deep Learning for Multimedia Processing)

Intelligent Automation & Soft Computing 2023, 37(3), 3235-3258. https://doi.org/10.32604/iasc.2023.041392

Received 20 April 2023; Accepted 08 June 2023; Issue published 11 September 2023

Abstract

Respiratory infections in children increase the risk of fatal lung disease, making effective identification and analysis of breath sounds essential. However, most studies have focused on adults ignoring pediatric patients whose lungs are more vulnerable due to an imperfect immune system, and the scarcity of medical data has limited the development of deep learning methods toward reliability and high classification accuracy. In this work, we collected three types of breath sounds from children with normal (120 recordings), bronchitis (120 recordings), and pneumonia (120 recordings) at the posterior chest position using an off-the-shelf 3M electronic stethoscope. Three features were extracted from the wavelet denoised signal: spectrogram, mel-frequency cepstral coefficients (MFCCs), and Delta MFCCs. The recognition model is based on transfer learning techniques and combines fine-tuned MobileNetV2 and modified ResNet50 to classify breath sounds, along with software for displaying analysis results. Extensive experiments on a real dataset demonstrate the effectiveness and superior performance of the proposed model, with average accuracy, precision, recall, specificity and F1 scores of 97.96%, 97.83%, 97.89%, 98.89% and 0.98, respectively, achieving superior performance with a small dataset. The proposed detection system, with a high-performance model and software, can help parents perform lung screening at home and also has the potential for a vast screening of children for lung disease.

---

Keywords

---