{
  "abstract": "Introduction Smartphone-based acoustic screening offers a low-cost, non-invasive approach for identifying sleep-disordered breathing (SDB) in home environments. Paediatric populations, however, present unique challenges, such as lower tolerance for sensors and weaker breathing sounds during sleep. This study reports early findings from a customised data collection and machine learning pipeline for screening paediatric SDB ( figure 1).Methods Audio was recorded using a smartphone while a simultaneous home sleep test (HST) was conducted on six children (ages 2-11) across both sleep lab and home settings. An optional thorax sensor was offered to record respiratory effort for future model development. Audio quality was assessed using signal-to-noise ratio (SNR), calculated as the snore-to-non-snore ratio. HST data were manually scored, and quality was evaluated using the percentage of missing oxygen saturation (SpO 2) data. Our existing acoustic model for SDB screening, pre-trained on adult data, was first evaluated, then fine-tuned and re-evaluated on the paediatric dataset using cross-validation.Results Data quality varied markedly by setting. Home recordings showed more frequent sensor dropout (4%-18% missing SpO 2) and greater SNR variability (-3 to 8 dB) than lab recordings (0%-2.64% missing SpO2; higher and more consistent SNR). Only two of six children tolerated the thorax sensor throughout the night. The adult acoustic model alone performed poorly on paediatric data, but fine-tuning with paediatric recordings improved AHI prediction accuracy (table 1).Discussion Initial findings confirm low tolerance of wearable sensors by children. Parents reported stress associated with lab-based studies and expressed strong preference for non-invasive, home-based alternatives. The smartphone app was consistently described as easy to use. Fine-tuned acoustic models showed promising improvements in estimating AHI from children’s sleep sounds. These findings support the feasibility and acceptability of acoustic screening in children and highlight the need for sensor-free, AI-driven diagnostic tools in paediatric sleep medicine.Abstract P57 Table 1Comparison of model performance across patients under various conditions. Missing SpO2 indicates the percentage of the recording without oxygen saturation data. SNR reflects signal quality, where higher values indicate clearer snore signals. Scored AHI is based on home sleep tests scored by qualified sleep technologists. FT indicates whether the model is finetuned on child’s dataAbstract P57 Figure 1Pipeline of data collection and machine learning model development. Breathing sounds are recorded using a smartphone application. Home sleep test (HST) data is annotated by clinicians to identify apnoea-hypopnea events and compute the apnoea-hypopnea index (AHI), which serves as the ground truth (referred to as ‘Scored AHI’) for training and evaluating the machine learning model",
  "authors": [
    {
      "affiliations": [
        "School of Computer Science, The University of Sheffield, Sheffield, United Kingdom"
      ],
      "name": "Veronica Rowe"
    },
    {
      "affiliations": [
        "School of Computer Science, The University of Sheffield, Sheffield, United Kingdom"
      ],
      "name": "Chaoyue Niu"
    },
    {
      "affiliations": [
        "School of Computer Science, The University of Sheffield, Sheffield, United Kingdom"
      ],
      "name": "Guy J Brown"
    },
    {
      "affiliations": [
        "Sheffield Children’s NHS Foundation Trust, Sheffield, United Kingdom"
      ],
      "name": "Heather Elphick"
    },
    {
      "affiliations": [
        "Sheffield Children’s NHS Foundation Trust, Sheffield, United Kingdom"
      ],
      "name": "Lowri Thomas"
    },
    {
      "affiliations": [
        "Passion for Life Healthcare, Chester, United Kingdom"
      ],
      "name": "Sam Johnson"
    },
    {
      "affiliations": [
        "School of Computer Science, The University of Sheffield, Sheffield, United Kingdom"
      ],
      "name": "Ning Ma"
    }
  ],
  "title": "P57 Smartphone-based acoustic screening for paediatric sleep disordered breathing in hospital and home settings: data collection and preliminary machine learning results",
  "uid": "84daecb5-6f1c-510a-83d9-b83eed9db2b0"
}