Smartphone AI: Can It Detect Abnormal Lung Sounds

How can AI help smartphones detect abnormal lung sounds?

A study published on July 30, 2026, in IEEE Journal of Biomedical and Health Informatics combined recordings from electronic stethoscopes and smartphones. Paired recordings helped researchers identify how the devices captured sounds differently. Their adapted model outperformed models trained exclusively on smartphone recordings. [Original study](https://pubmed.ncbi.nlm.nih.gov/42531109/).

This addresses a practical obstacle in digital medicine: a system trained with one recording device may perform differently when another captures the signal. The researchers used audio preprocessing and an adaptation method to help the model transfer information from stethoscope recordings to smartphone audio. The intended application is remote respiratory monitoring. [Study methods](https://pubmed.ncbi.nlm.nih.gov/42531109/).

How reliable are the results for children's respiratory care?

The model achieved an area under the receiver operating characteristic curve, or AUROC, of 0.911 in the primary dataset and 0.862 in the external cohort. [Study results](https://pubmed.ncbi.nlm.nih.gov/42531109/). AUROC describes how well a model separates categories across possible decision thresholds; it should not be read as the percentage of children correctly diagnosed. For practical care, clinicians also need to know how often a chosen threshold misses abnormal sounds or raises false alarms.

A 2025 systematic review in the Journal of Medical Internet Research examined 41 studies of pediatric lung-sound AI. It found encouraging classification performance alongside substantial differences in datasets, labeling methods, and evaluation approaches. Many studies used small datasets from individual centers. These limitations make larger, independent evaluations essential before assuming that results will transfer reliably between clinics or devices. [Systematic review](https://www.jmir.org/2025/1/e66491/).

The clinical distinction is equally important: recognizing an abnormal sound is one part of assessing a child. It does not automatically identify the illness causing that sound or determine treatment. The review distinguished studies that classified sounds from those that attempted to classify diseases. The practical implication is that an acoustic result should support a broader assessment, with its intended diagnostic role tested explicitly. [Review of classification tasks](https://www.jmir.org/2025/1/e66491/).

What would smartphone lung monitoring mean for families?

A separate study published in the European Journal of Pediatrics in May 2026 illustrates the remaining recording challenge. Researchers collected 2,020 smartphone recordings from 217 children, but only 1,500 recordings met quality criteria. Their wheeze-detection model was evaluated on those usable recordings. This means performance on selected audio cannot automatically describe how a system would behave whenever a caregiver presses record at home. The authors called for further optimization and multicenter validation. [Pediatric smartphone auscultation study](https://pmc.ncbi.nlm.nih.gov/articles/PMC13167856/).

For future services, an important design requirement would be a clear response when audio is unusable, alongside instructions about when to seek care. Families should never wait for an app result when a child is struggling to breathe. NHS guidance identifies breathing pauses, blue or grey lips or tongue, and difficulty waking as emergency warning signs. Call the local emergency number if these occur. Poor feeding, dehydration, or worsening illness also warrant prompt medical advice. [NHS respiratory warning signs](https://www.nhs.uk/conditions/bronchiolitis/).