Classification of lung sounds using scalogram representation of sound segments and convolutional neural network.

Lung auscultation is one of the most common methods for screening of lung diseases. The increasingly high rate of respiratory diseases leads to the need for robust methods to detect the abnormalities in patients' breathing sounds. Lung sounds analysis stands out as a promising approach to automatic screening of lung diseases, serving as a second opinion for doctors as a stand-alone device for preliminary screening of lung diseases in remote areas. In previous research on lung classification using ICBHI Database on Kaggle, lung audios are converted to spectral images and fed into deep neural networks for training. There are a few studies which uses the scalogram, however they focussed on classification among different lung diseases. The use of scalograms in categorising the sound types are rarely used. In this paper, we combined scalograms and neural networks for classification of lung sound types. Padding methods and augmentation are also considered to evaluate the impacts on classification score. An ensemble learning is incorporated to increase classification accuracy by utilising voting of many models. The model trained and evaluated has shown prominent improvement of this method on classification on the benchmark ICBHI database.

Evaluation of the Luminex xTAG Respiratory Viral Panel FAST v2 assay for detection of multiple respiratory viral pathogens in nasal and throat swabs in Vietnam.

BACKGROUND: Acute respiratory infections (ARI) are among the leading causes of hospitalization in children ≤5 years old. Rapid diagnostics of viral pathogens is essential to avoid unnecessary antibiotic treatment, thereby slowing down antibiotic-resistance. We evaluated the diagnostic performance of the Luminex xTAG Respiratory Viral Panel FAST v2 against viral specific PCR as reference assays for ARI in Vietnam. METHODS: Four hundred and forty two nose and throat swabs were collected in viral transport medium, and were tested with Luminex xTAG Respiratory Viral Panel FAST v2. Multiplex RT-PCR and single RT-PCR were used as references.    Results: Overall, viral pathogens were detected in a total count of 270/294 (91.8%, 95% CI 88.1-94.7) by the Luminex among reference assays, whilst 112/6336 (1.8%, 95% CI, 1.4-2.1) of pathogens were detected by the Luminex, but not by reference assays. Frequency of pathogens detected by Luminex and reference assays was 379 and 292, respectively. The diagnostic yield was 66.7% (295/442, 95%CI 62.1-71.1%) for the Luminex assay and 54.1% (239/442, 95% CI, 49.3-58.8%) for reference assays. The Luminex kit had higher yields for all viruses except influenza B virus, respiratory syncytial virus, and human bocavirus. High agreements between both methods [mean (range): 0.91 (0.83-1.00)] were found for 10/15 viral agents. CONCLUSIONS: The Luminex assay is a high throughput multiplex platform for rapid detection of common viral pathogens causing ARI. Although the current high cost may prevent Luminex assays from being widely used, especially in limited resource settings where ARI are felt most, its introduction in clinical diagnostics may help reduce unnecessary use of antibiotic prescription.