Deep Learning-Based Venous Gas Emboli Grade Classification in Doppler Ultrasound Audio Recordings

Azarang, Arian; Le, David; Hoang, Andrew H; Blogg, Lesley S; Dayton, Paul A.; Lance, Rachel M.; Natoli, Michael J.; Gatrell, Alan; Tillmans, Frauke; Moon, Richard E.; Lindholm, Peter; Papadopoulou, Virginie

doi:10.1109/tbme.2022.3217711

Cited by 2 publications

(5 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This factor may cause the reduced amount of air released during stent-graft deployment when compared to a saline solution ( 205 , 207 ). Recent research combining artificial intelligence with Doppler ultrasound recordings for vascular air embolism detection and grading suggests promise for combining traditional and emerging technologies for improved diagnostic and bedside care in the area of air embolism ( 4 , 233 ). Antiquated yet of paramount importance, simple caution and alertness may allow surgeons and anesthesiologists to see or hear air being aspirated into circulation during the case while simultaneously reviewing the multi- technological monitoring devices for confirmation of the presence of a vascular air embolism ( 4 ).…”

Section: Prevention and Early Diagnosis Of Vascular Air Embolism: Cli...mentioning

confidence: 99%

Iatrogenic air embolism: pathoanatomy, thromboinflammation, endotheliopathy, and therapies

Marsh,

Moore,

Moore

et al. 2023

Front. Immunol.

View full text Add to dashboard Cite

Iatrogenic vascular air embolism is a relatively infrequent event but is associated with significant morbidity and mortality. These emboli can arise in many clinical settings such as neurosurgery, cardiac surgery, and liver transplantation, but more recently, endoscopy, hemodialysis, thoracentesis, tissue biopsy, angiography, and central and peripheral venous access and removal have overtaken surgery and trauma as significant causes of vascular air embolism. The true incidence may be greater since many of these air emboli are asymptomatic and frequently go undiagnosed or unreported. Due to the rarity of vascular air embolism and because of the many manifestations, diagnoses can be difficult and require immediate therapeutic intervention. An iatrogenic air embolism can result in both venous and arterial emboli whose anatomic locations dictate the clinical course. Most clinically significant iatrogenic air emboli are caused by arterial obstruction of small vessels because the pulmonary gas exchange filters the more frequent, smaller volume bubbles that gain access to the venous circulation. However, there is a subset of patients with venous air emboli caused by larger volumes of air who present with more protean manifestations. There have been significant gains in the understanding of the interactions of fluid dynamics, hemostasis, and inflammation caused by air emboli due to in vitro and in vivo studies on flow dynamics of bubbles in small vessels. Intensive research regarding the thromboinflammatory changes at the level of the endothelium has been described recently. The obstruction of vessels by air emboli causes immediate pathoanatomic and immunologic and thromboinflammatory responses at the level of the endothelium. In this review, we describe those immunologic and thromboinflammatory responses at the level of the endothelium as well as evaluate traditional and novel forms of therapy for this rare and often unrecognized clinical condition.

show abstract

Section: Prevention and Early Diagnosis Of Vascular Air Embolism: Cli...mentioning

confidence: 99%

Iatrogenic air embolism: pathoanatomy, thromboinflammation, endotheliopathy, and therapies

Marsh,

Moore,

Moore

et al. 2023

Front. Immunol.

View full text Add to dashboard Cite

show abstract

“…Additionally, compared to the sole use of Spencer grade classification accuracy as the final metric for DU VGE analysis, the ground truth bubble-only DU data containing the exact number and placement of bubbles is available to researchers, and can be used for more detailed evaluation metrics. For example, a subset of the synthetic data presented here has been used in our prior work, enhancing the performance of a deep-learning algorithm tasked with Spencer grade classifications [ 23 ]. Additional validation of this dataset as a benchmark for signal separation or other VGE classification systems could be envisaged in future work.…”

Section: Discussionmentioning

confidence: 99%

“…The method for generating and collecting single-bubble Doppler recordings is described elsewhere [ 23 ]. Briefly described, a tissue-mimicking flow phantom was fabricated by first dissolving 85% distilled water with 10g/mL porcine gelatin (Gel strength 300, type-A, Sigma-Aldrich, St. Louis, MO, USA), and 5% 1-propanol (Fisher Chemical, Hampton, NH, USA).…”

Section: Methodsmentioning

confidence: 99%

“…Synthetic post-dive DU data was generated using the following process, originally proposed in Azarang et al 2022 [ 23 ], but is expanded to include subclavian data as well as selective placement of VGE in between cardiac cycles without cardiac interference. We initially select one of the available human baseline recordings, and then segment the audio to the desired length with an added 50% sample buffer at a random location along the signal.…”

Section: Methodsmentioning

confidence: 99%

“…However, cardiac sounds from precordial DU as well as bubble signals have not been modeled making direct simulations of post-decompression data not feasible. In our previous work, we developed a novel deep learning approach for automated Doppler classification trained initially on synthetic data and fine-tuned on real-data [ 23 ]. Due to the complexities in fully simulating relevant and accurate synthetic DU data, we procedurally combined experimentally collected human cardiac and flowing bubble DU audio creating data spanning all 5 Spencer grades.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

An open-source framework for synthetic post-dive Doppler ultrasound audio generation

Hoang

Azarang

et al. 2023

PLoS ONE

View full text Add to dashboard Cite

Doppler ultrasound (DU) measurements are used to detect and evaluate venous gas emboli (VGE) formed after decompression. Automated methodologies for assessing VGE presence using signal processing have been developed on varying real-world datasets of limited size and without ground truth values preventing objective evaluation. We develop and report a method to generate synthetic post-dive data using DU signals collected in both precordium and subclavian vein with varying degrees of bubbling matching field-standard grading metrics. This method is adaptable, modifiable, and reproducible, allowing for researchers to tune the produced dataset for their desired purpose. We provide the baseline Doppler recordings and code required to generate synthetic data for researchers to reproduce our work and improve upon it. We also provide a set of pre-made synthetic post-dive DU data spanning six scenarios representing the Spencer and Kisman-Masurel (KM) grading scales as well as precordial and subclavian DU recordings. By providing a method for synthetic post-dive DU data generation, we aim to improve and accelerate the development of signal processing techniques for VGE analysis in Doppler ultrasound.

show abstract

Deep Learning-Based Venous Gas Emboli Grade Classification in Doppler Ultrasound Audio Recordings

Cited by 2 publications

References 33 publications

Iatrogenic air embolism: pathoanatomy, thromboinflammation, endotheliopathy, and therapies

Iatrogenic air embolism: pathoanatomy, thromboinflammation, endotheliopathy, and therapies

An open-source framework for synthetic post-dive Doppler ultrasound audio generation

Contact Info

Product

Resources

About