An accurate deep learning model for wheezing in children using real world data

Abstract

Auscultation is an important diagnostic method for lung diseases. However, it is a subjective modality and requires a high degree of expertise. To overcome this constraint, artificial intelligence models are being developed. However, these models require performance improvements and do not reflect the actual clinical situation. We aimed to develop an improved deep-learning model learning to detect wheezing in children, based on data from real clinical practice. In this prospective study, pediatric pulmonologists recorded and verified respiratory sounds in 76 pediatric patients who visited a university hospital in South Korea. In addition, structured data, such as sex, age, and auscultation location, were collected. Using our dataset, we implemented an optimal model by transforming it based on the convolutional neural network model. Finally, we proposed a model using a 34-layer residual network with the convolutional block attention module for audio data and multilayer perceptron layers for tabular data. The proposed model had an accuracy of 91.2%, area under the curve of 89.1%, precision of 94.4%, recall of 81%, and F1-score of 87.2%. The deep-learning model proposed had a high accuracy for detecting wheeze sounds. This high-performance model will be helpful for the accurate diagnosis of respiratory diseases in actual clinical practice.

An Accelerometer-Based Wearable Patch for Robust Respiratory Rate and Wheeze Detection Using Deep Learning

“…After the physician labeled the mel spectrograms of each recording segment, the accelerometer patch time–frequency wheeze detection and 2D CNN wheeze detection were benchmarked with each other as followed in Figure 5 c.i and Figure 5 c.ii, respectively. The deterministic time–frequency analysis and the deep learning method are recognized as the standard for objective wheeze detection in other studies on digital stethoscope recordings [ 32 , 40 , 41 , 42 , 48 ]. We aim to benchmark the two established concepts with the PIV mel spectrogram collected from the accelerometer patch for accuracy, sensitivity, and specificity to determine the optimal approach for wheeze detection.…”

“…One popular deep learning model for image classification is CNN [ 40 , 41 , 50 , 51 , 52 ]. Since the captured PIVs were represented as mel spectrogram images, the accelerometer patch mel spectrograms can be used as inputs for this deep learning model.…”

“…To circumvent these problems, a promising, novel approach is to capture only the acoustic vibrations of the chest walls using a chip-scale wideband accelerometer, which acts as a contact microphone by measuring the vibrations of a surface that it is mounted on while also isolating from ambient sounds [ 29 , 30 , 34 , 35 , 36 , 37 , 38 , 39 ]. This is used in a miniature wearable patch and deploys objective and validated methods, like computation detection, to detect wheezing [ 32 , 40 , 41 , 42 , 43 ].…”

Oxidation and phase transfer of individual Cr-doped dendritic FeOx particles visualized by full-field nano-XAFS spectroimaging

“…Combination of X-ray imaging techniques, such as scanning transmission/fluorescence X-ray microscopy (STXM/SFXM), [17][18][19][20][21][22][23] full-field transmission X-ray microscopy (FF-TXM), [24][25][26][27][28][29][30][31][32][33][34] coherent X-ray diffraction imaging (CXDI) including X-ray ptychography, [35][36][37][38][39][40][41][42] with X-ray absorption spectroscopy, has been employed to visualize the chemical state distribution in materials. 43 STXM/SFXM or scanning nano-XAFS uses X-ray beams focused by a Fresnel zone plate (FZP) or Kirkpatrick-Baez mirrors on the submicro to nanoscale to obtain X-ray transmission/fluorescence and chemical state images of materials.…”

Transfer Learning based Diagnosis and Analysis of Lung Sound Aberrations

“…In recent years, lung sound diagnosis and classification have attracted much research attention [1][2][3][4][5][6][7]. Breathing is so necessary that in 24 hours, an average human can breathe 25,000 times [8][9][10][11][12]. According to the World Health Organization (WHO), the COVID-19 epidemic on May 24, 2021, there have been 166,860,081 verified cases, with 3,459,996 deaths recorded [13].…”