AI Tool Improves Smartphone Detection of Abnormal Pediatric Lung Sounds

Respiratory disease in children remains a major clinical concern worldwide. Abnormal lung sounds can indicate pediatric respiratory illness, but standard auscultation usually requires an in-person examination with a stethoscope. This limits access to screening between visits and in remote settings. Researchers have now developed an artificial intelligence framework that adapts electronic stethoscope training data for smartphone-based detection of abnormal pediatric lung sounds.

The framework was developed by a team in the Department of Applied Artificial Intelligence at Seoul National University of Science and Technology in South Korea. It is designed for smartphone deployment and addresses the acoustic mismatch between electronic stethoscope recordings and smartphone microphone audio. The study was published in the IEEE Journal of Biomedical and Health Informatics on October 16, 2026.

The researchers collected paired lung sound recordings using electronic stethoscopes and smartphones to guide adaptation between the two devices. The primary dataset included electronic stethoscope recordings from 611 children aged 0–6 years, including 147 who also provided paired smartphone recordings collected at the same four anatomical sites. An independent external validation dataset included smartphone-only recordings from 48 children.

To align the two recording formats, smartphone audio was processed with a 1 kHz low-pass filter and downsampled from 48 kHz to 4 kHz. The model also incorporated a directional MixStyle module that transferred smartphone-domain feature characteristics into the stethoscope domain during training. This approach was integrated into an Audio Spectrogram Transformer deep learning model for lung sound classification.

On smartphone recordings in the primary dataset, the model achieved an F1 score of 0.802 and an area under the receiver operating characteristic curve (AUROC) of 0.911 for detecting abnormal lung sounds. By comparison, a smartphone-only model achieved an F1 score of 0.732, while naïvely pooling stethoscope and smartphone recordings produced a score of 0.708.

Performance was also maintained in the independent smartphone cohort. A model trained only on stethoscope data achieved an F1 score of 0.034 and an AUROC of 0.388, while the proposed device-adaptation model reached an F1 score of 0.744 and an AUROC of 0.862.

The findings show that signal-level alignment can reduce acoustic discrepancies between electronic stethoscopes and smartphones. The framework may support more accessible respiratory screening and monitoring, but larger multisite cohorts, real-world home recordings, and testing across multiple smartphone models are still needed before clinical deployment.

“To a neural network, the smartphone and electronic stethoscope effectively speak different languages. Instead of simply pooling the data from the two devices, our approach offers reliable domain adaptation, where the model learns from reliable labeled electronic stethoscope recordings while adapting to the smartphone domain,” said Seong-Eun Kim, associate professor in the Department of Applied Artificial Intelligence at Seoul National University of Science and Technology.

“Our framework provides a proof of concept for scalable and cost-effective pediatric respiratory screening. It could help parents screen children at home between visits while keeping clinicians in the loop. Moreover, such technology could improve remote clinic consultations and aid community health care workers in frontline triage in remote areas,” added Kim.

Related Links
Seoul National University of Science and Technology