Development of a Deep Learning Network to Classify Inferior Vena Cava Collapse to Predict Fluid Responsiveness

Abstract

Objectives-To create a deep learning algorithm capable of video classification, using a long short-term memory (LSTM) network, to analyze collapsibility of the inferior vena cava (IVC) to predict fluid responsiveness in critically ill patients.Methods-We used a data set of IVC ultrasound (US) videos to train the LSTM network. The data set was created from IVC US videos of spontaneously breathing critically ill patients undergoing intravenous fluid resuscitation as part of 2 prior prospective studies. We randomly selected 90% of the IVC videos to train the LSTM network and 10% of the videos to test the LSTM network's ability to predict fluid responsiveness. Fluid responsiveness was defined as a greater than 10% increase in the cardiac index after a 500-mL fluid bolus, as measured by bioreactance.Results-We analyzed 211 videos from 175 critically ill patients: 191 to train the LSTM network and 20 to test it. Using standard data augmentation techniques, we increased our sample size from 191 to 3820 videos. Of the 175 patients, 91 (52%) were fluid responders. The LSTM network was able to predict fluid responsiveness moderately well, with an area under the receiver operating characteristic curve of 0.70 (95% confidence interval [CI], 0.43-1.00), a positive likelihood ratio of infinity, and a negative likelihood ratio of 0.3 (95% CI, 0.12-0.77). In comparison, point-of-care US experts using video review offline and manual diameter measurement via software caliper tools achieved an area under the receiver operating characteristic curve of 0.94 (95% CI, 0.83-0.99).Conclusions-We demonstrated that an LSTM network can be trained by using videos of IVC US to classify IVC collapse to predict fluid responsiveness. Our LSTM network performed moderately well given the small training cohort but worse than point-of-care US experts. Further training and testing of the LSTM network with a larger data sets is warranted.

Diagnostic Accuracy of Ultrasonographic Respiratory Variation in the Inferior Vena Cava, Subclavian Vein, Internal Jugular Vein, and Femoral Vein Diameter to Predict Fluid Responsiveness: A Systematic Review and Meta-Analysis

“…After the full-text review, 62 articles were excluded due to insufficient data ( n = 36), lack of 2-by-2 data ( n = 25), and not being original ( n = 1). Finally, 30 studies [ 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 ] comprising 1719 patients were included in this review ( Figure 1 ); detailed information about the eligible studies is shown in Table 1 . In cases of the IJV [ 29 , 36 , 42 ], FV (was not detected), and SCV [ 41 ], we were not able to conduct the meta-analysis due to an insufficient number of studies.…”

“…We extracted the results of experts because other studies were conducted by experts. One study by Blavius [ 50 ] was a comparative study between artificial intelligence and human. We extracted the result of the training dataset by humans because the number of test datasets was much smaller than the test set (20 vs. 175).…”

Advanced Ultrasound and Photoacoustic Imaging in Cardiology

“…They used a public dataset with 750,018 individual ultrasound images of five different types and showed that the classification accuracy varied from 96% to 85.6% for the various models, with VGG-16 giving the best performance while the DenseNet201 performed the worst for classification. Another work by Blaivas et al [ 103 ] proposed a LSTM network for inferior vena cava (IVC) POCUS videos in patients undergoing the intravenous fluid resuscitation and use 211 videos and achieved the receiver operating characteristic curve of 0.70 (95% confidence interval [CI], 0.43–1.00) for predicting the fluid responsiveness. Generative Adversarial Networks (GANS) have also gained popularity for generating more data as well as applicable in the cases where the paired input/output pairs are not easily available for training the models.…”

Machine Learning and Precision Medicine in Emergency Medicine: The Basics

“…In a third study, Blaivas et al created a DL algorithm to analyze collapsibility of the inferior vena cava to predict fluid responsiveness in critically ill patients. The DL algorithm predicted fluid responsiveness with an AUC of 0.70, compared to an AUC of 0.94 by two POCUS experts using video review and manual caliper measurements [ 27 ].…”

“…DL has also been applied to help make more advanced diagnoses including heart failure with preserved ejection fraction, cardiomyopathy, amyloidosis, and pulmonary hypertension [20][21][22][23][24]. [27].…”