The full article is currently under embargo for 12 months and will be made available on our website and PMC after this period. If you would like to access the article before it becomes publicly available, you can purchase it for personal use directly on our website. Alternatively, you may join SPUMS or EUBS to gain access to the entire issue.
An echo from the past: open access repository of over 10,000 annotated Doppler audio recordings of venous gas emboli
S Lesley Blogg1,2, Arian Azarang3, Rachel Lance4,5, Frauke Tillmans4,6, Richard Moon5, Virginie Papadopoulou3, Peter Lindholm1,7
1 Department of Emergency Medicine, University of California San Diego, USA
2 SLB Consulting, c/o Home Park Barn, Kirkby Stephen, Cumbria, UK
3 Department of Radiology, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA
4 Divers Alert Network, Durham, North Carolina, USA
5 Departments of Anesthesiology and Medicine, Center for Hyperbaric Medicine and Environmental Physiology, Duke University, Durham, North Carolina, USA
6 Lampe Joint Department of Biomedical Engineering, The University of North Carolina at Chapel Hill and North Carolina State University, Chapel Hill, North Carolina, USA
7 Department of Physiology and Pharmacology, Karolinska Institute, Sweden
Corresponding author: Professor Peter Lindholm, 200 W Arbor Drive, MC8676, San Diego, CA 92103, USA
Keywords
Artificial intelligence; Bubbles; Decompression sickness; Measurement; Venous gas embolism
Abstract
(Blogg SL, Azarang A, Lance R, Tillmans F, Moon R, Papadopolou V, Lindholm P. An echo from the past: open access repository of over 10,000 annotated Doppler audio recordings of venous gas emboli. Diving and Hyperbaric Medicine. 2026 31 March;56(1):8−12. doi: 10.28920/dhm56.1.8-12. PMID: 41875437.)
Introduction: Doppler ultrasound measurements have been recorded since the 1970s across the world and provide a valuable data resource for learning, analysis, and potential training of deep learning algorithms to recognise and grade venous gas emboli (VGE) allowing assessment of decompression sickness (DCS) risk.
Methods: We collected a ‘big database’ of Doppler recordings and associated metadata. Audio tapes with recorded Doppler data were converted to digital files, then cut into individual recordings and matched with their metadata, including subject and pressure profile information. The audio signals and their Doppler grades were then processed further for suitability to train an algorithm to identify VGE.
Results: A total of 10,099 Doppler ultrasound recordings were compiled. Divers (n = ≤ 311; 170 identified, ≤ 141 unidentified) were male, with a median age of 31.5 years among the 170 identified divers. The maximum depth of the dives included ranged from 24 m (80 feet) to 91.4 m (300 feet). The timing of the Doppler measurements ranged from two minutes post-dive to 594 min post-dive, with a median time of 52 min. Breathing gases included air, nitrox, and heliox. DCS was noted in only 12 individuals. The dataset centred around lower VGE loads (Spencer Grades 0, I, and II).
Conclusions: This database represents a landmark in DCS investigation as the audio dataset and metadata collected have been released under a public domain license for further use. The large number of data points has also allowed the development of a deep learning algorithm that can grade bubble loads without a human operator.
Copyright: This article is the copyright of the authors who grant Diving and Hyperbaric Medicine a non-exclusive licence to publish the article in electronic and other forms.
