Meningitis Screening in Young Infants Based on a Novel, Non- Invasive, Transfontanellar Ultrasound Device: a Proof-of- Concept Study

Abstract

Abstract Background Meningitis is a potentially life-threatening disease if not promptly diagnosed and treated. Clinical presentation is often unspecific, especially among young infants and newborns, justifying the need to perform lumbar punctures (LP) to obtain cerebrospinal fluid (CSF) for a laboratory-based confirmation. In high-income settings, LPs are often part of the protocolized systematic approach to screen for meningitis. Given its relatively low incidence, most are not confirmatory. The aim of this study was to validate a novel transfontanellar ultrasound-based technique to screen for meningitis, designed to non-invasively identify ranges of white blood cells (WBC) in CSF, to be used on patients with criteria for a LP. Methods We prospectively recruited patients under one year of age, with suspected meningitis, a permeable fontanelle and a LP performed within 24h before enrolment, from three Spanish University Hospitals (2021–2023). Images showing the backscatter pattern from CSF were obtained using a customized high-resolution (HR) ultrasonic probe. A deep-learning model (DL) was trained to classify CSF patterns according to WBC values obtained through the LP, setting a 30 cells/mm3 threshold to differentiate controls from cases. Results We obtained a set of 781 HR CSF images from 16 targeted and recruited patients. In parallel, 17 LPs were performed, confirming 6 meningitis cases (one patient had a second LP to verify response to treatment) which were paired to 445 CSF frames, and 10 controls, with 336 CSF images. After training the DL model, the device showed a sensitivity of 100% and a specificity of 90% at patients' level, with one control misclassified. Conclusion This proof-of-concept study confirmed that our device, based on ultrasound and DL, could potentially be used as an automated screening method to modulate indications to perform LPs.