Validation of machine learning models for estimation of left ventricular ejection fraction on point-of-care ultrasound: insights on features that impact performance.

BACKGROUND: Machine learning (ML) algorithms can accurately estimate left ventricular ejection fraction (LVEF) from echocardiography, but their performance on cardiac point-of-care ultrasound (POCUS) is not well understood. OBJECTIVES: We evaluate the performance of an ML model for estimation of LVEF on cardiac POCUS compared with Level III echocardiographers' interpretation and formal echo reported LVEF. METHODS: Clinicians at a tertiary care heart failure clinic prospectively scanned 138 participants using hand-carried devices. Video data were analyzed offline by an ML model for LVEF. We compared the ML model's performance with Level III echocardiographers' interpretation and echo reported LVEF. RESULTS: There were 138 participants scanned, yielding 1257 videos. The ML model generated LVEF predictions on 341 videos. We observed a good intraclass correlation (ICC) between the ML model's predictions and the reference standards (ICC = 0.77-0.84). When comparing LVEF estimates for randomized single POCUS videos, the ICC between the ML model and Level III echocardiographers' estimates was 0.772, and it was 0.778 for videos where quantitative LVEF was feasible. When the Level III echocardiographer reviewed all POCUS videos for a participant, the ICC improved to 0.794 and 0.843 when only accounting for studies that could be segmented. The ML model's LVEF estimates also correlated well with LVEF derived from formal echocardiogram reports (ICC = 0.798). CONCLUSION: Our results suggest that clinician-driven cardiac POCUS produces ML model LVEF estimates that correlate well with expert interpretation and echo reported LVEF.

Automated Atrial Fibrillation Diagnosis by Echocardiography without ECG: Accuracy and Applications of a New Deep Learning Approach.

BACKGROUND: Automated rhythm detection on echocardiography through artificial intelligence (AI) has yet to be fully realized. We propose an AI model trained to identify atrial fibrillation (AF) using apical 4-chamber (AP4) cines without requiring electrocardiogram (ECG) data. METHODS: Transthoracic echocardiography studies of consecutive patients ≥ 18 years old at our tertiary care centre were retrospectively reviewed for AF and sinus rhythm. The study was first interpreted by level III-trained echocardiography cardiologists as the gold standard for rhythm diagnosis based on ECG rhythm strip and imaging assessment, which was also verified with a 12-lead ECG around the time of the study. AP4 cines with three cardiac cycles were then extracted from these studies with the rhythm strip and Doppler information removed and introduced to the deep learning model ResNet(2+1)D with an 80:10:10 training-validation-test split ratio. RESULTS: 634 patient studies (1205 cines) were included. After training, the AI model achieved high accuracy on validation for detection of both AF and sinus rhythm (mean F1-score = 0.92; AUROC = 0.95). Performance was consistent on the test dataset (mean F1-score = 0.94, AUROC = 0.98) when using the cardiologist's assessment of the ECG rhythm strip as the gold standard, who had access to the full study and external ECG data, while the AI model did not. CONCLUSIONS: AF detection by AI on echocardiography without ECG appears accurate when compared to an echocardiography cardiologist's assessment of the ECG rhythm strip as the gold standard. This has potential clinical implications in point-of-care ultrasound and stroke risk stratification.

Machine learning derived echocardiographic image quality in patients with left ventricular systolic dysfunction: insights on the echo views of greatest image quality.

We sought to determine the cardiac ultrasound view of greatest quality using a machine learning (ML) approach on a cohort of transthoracic echocardiograms (TTE) with abnormal left ventricular (LV) systolic function. We utilize an ML model to determine the TTE view of highest quality when scanned by sonographers. A random sample of TTEs with reported LV dysfunction from 09/25/2017-01/15/2019 were downloaded from the regional database. Component video files were analyzed using ML models that jointly classified view and image quality. The model consisted of convolutional layers for extracting spatial features and Long Short-term Memory units to temporally aggregate the frame-wise spatial embeddings. We report the view-specific quality scores for each TTE. Pair-wise comparisons amongst views were performed with Wilcoxon signed-rank test. Of 1,145 TTEs analyzed by the ML model, 74.5% were from males and mean LV ejection fraction was 43.1 ± 9.9%. Maximum quality score was best for the apical 4 chamber (AP4) view (70.6 ± 13.9%, p<0.001 compared to all other views) and worst for the apical 2 chamber (AP2) view (60.4 ± 15.4%, p<0.001 for all views except parasternal short-axis view at mitral/papillary muscle level, PSAX M/PM). In TTEs scanned by professional sonographers, the view with greatest ML-derived quality was the AP4 view.

A Novel Continuous Left Ventricular Diastolic Function Score Using Machine Learning.

BACKGROUND: Unlike left ventricular (LV) ejection fraction, which provides a precise, reliable, and prognostically valuable measure of systolic function, there is no single analogous measure of LV diastolic function. OBJECTIVES: We aimed to develop a continuous score to grade LV diastolic function using machine learning modeling of echocardiographic data. METHODS: Consecutive echo studies performed at a tertiary-care center between February 1, 2010, and March 31, 2016, were assessed, excluding studies containing features that would interfere with diastolic function assessment as well as studies in which 1 or more parameters within the contemporary diastolic function assessment algorithm were not reported. Diastolic function was graded based on 2016 American Society of Echocardiography (ASE)/European Association of Cardiovascular Imaging (EACVI) guidelines, excluding indeterminate studies. Machine learning models were trained (support vector machine [SVM], decision tree [DT], XGBoost [XGB], and dense neural network [DNN]) to classify studies within the training set by diastolic dysfunction severity, blinded to the ASE/EACVI classification. The DNN model was retrained to generate a regression model (R-DNN) to predict a continuous LV diastolic function score. RESULTS: A total of 28,986 studies were included; 23,188 studies were used to train the models, and 5,798 studies were used for validation. The models were able to reclassify studies with high agreement to the ASE/EACVI algorithm (SVM, 83%; DT, 100%; XGB, 100%; DNN, 98%). The continuous diastolic function score corresponded well with ASE/EACVI guidelines, with scores of 1.00 ± 0.01 for studies with normal function and 0.74 ± 0.05, 0.51 ± 0.06, and 0.27 ± 0.11 for mild, moderate, and severe diastolic dysfunction, respectively (mean ± 1 SD). A score of <0.91 predicted abnormal diastolic function (area under the receiver operator curve = 0.99), while a score of <0.65 predicted elevated filling pressure (area under the receiver operator curve = 0.99). CONCLUSIONS: Machine learning can assimilate echocardiographic data and generate an automated continuous diastolic function score that corresponds well with current diastolic function grading recommendations.

Impact of the updated diastolic function guidelines in the real world.

BACKGROUND: Echocardiographic assessment of diastolic function is complex but can aid in the diagnosis of heart failure, particularly in patients with preserved ejection fraction. In 2016, the American Society of Echocardiography (ASE) and European Association of Cardiovascular Imaging (EACVI) published an updated algorithm for the evaluation of diastolic function. The objective of our study was to assess its impact on diastolic function assessment in a real-world cohort of echo studies. METHODS: We retrospectively identified 71,727 consecutive transthoracic echo studies performed at a tertiary care center between February 2010 and March 2016 in which diastolic function was reported based on the 2009 ASE Guidelines. We then programmed a software algorithm to assess diastolic function in these echo studies according to the 2016 ASE/EACVI Guidelines. RESULTS: When diastolic function assessment based on the 2009 guidelines was compared to that using the 2016 guidelines, there were significant differences in proportion of studies classified as normal (23% vs. 32%) or indeterminate (43% vs. 36%) function, and mild (23% vs. 23%), moderate (10% vs. 8%), or severe (1% vs. 2%) diastolic dysfunction, with poor agreement between the two methods (Kappa 0.323, 95% CI 0.318-0.328). Furthermore, within the subgroup of studies with preserved ejection fraction and no evidence of myocardial disease, there was significant reclassification from mild diastolic dysfunction to normal diastolic function. CONCLUSION: The updated guidelines result in significant differences in diastolic function interpretation in the real world. Our findings have important implications for the identification of patients with or at risk for heart failure.

Comparison of Patient Position and Midline Lumbar Neuraxial Access Via Statistical Model Registration to Ultrasound.

Patient positioning and needle puncture site are important for lumbar neuraxial anesthesia. We sought to identify optimal patient positioning and puncture sites with a novel ultrasound registration. We registered a statistical model to volumetric ultrasound data acquired from volunteers (n = 10) in three positions: (i) prone; (ii) seated with thoracic and lumbar flexion; and (iii) seated as in position ii, with a 10° dorsal tilt. We determined injection target size and penetration success by simulating lumbar injections on validated registered models. Injection window and target area sizes in seated positions were significantly larger than those in prone positions by 65% in L2-3 and 130% in L3-4; a 10° tilt had no significant effect on target sizes between seated positions. In agreement with computed tomography studies, simulated L2-3 and L3-4 injections had the highest success at the 50% and 75% midline puncture sites, respectively, measured from superior to inferior spinous process. We conclude that our registration to ultrasound technique is a potential tool for tolerable determination of puncture site success in vivo.

Cardiac Phase Detection in Echocardiograms With Densely Gated Recurrent Neural Networks and Global Extrema Loss.

Accurate detection of end-systolic (ES) and end-diastolic (ED) frames in an echocardiographic cine series can be difficult but necessary pre-processing step for the development of automatic systems to measure cardiac parameters. The detection task is challenging due to variations in cardiac anatomy and heart rate often associated with pathological conditions. We formulate this problem as a regression problem and propose several deep learning-based architectures that minimize a novel global extrema structured loss function to localize the ED and ES frames. The proposed architectures integrate convolution neural networks (CNNs)-based image feature extraction model and recurrent neural networks (RNNs) to model temporal dependencies between each frame in a sequence. We explore two CNN architectures: DenseNet and ResNet, and four RNN architectures: long short-term memory, bi-directional LSTM, gated recurrent unit (GRU), and Bi-GRU, and compare the performance of these models. The optimal deep learning model consists of a DenseNet and GRU trained with the proposed loss function. On average, we achieved 0.20 and 1.43 frame mismatch for the ED and ES frames, respectively, which are within reported inter-observer variability for the manual detection of these frames.

Model-based registration of preprocedure MR and intraprocedure US of the lumbar spine.

PURPOSE: Epidural and spinal needle insertions, as well as facet joint denervation and injections are widely performed procedures on the lumbar spine for delivering anesthesia and analgesia. Ultrasound (US)-based approaches have gained popularity for accurate needle placement, as they use a non-ionizing, inexpensive and accessible modality for guiding these procedures. However, due to the inherent difficulties in interpreting spinal US, they yet to become the clinical standard-of-care. METHODS: A novel statistical shape [Formula: see text] pose [Formula: see text] scale (s [Formula: see text] p [Formula: see text] s) model of the lumbar spine is jointly registered to preoperative magnetic resonance (MR) and US images. An instance of the model is created for each modality. The shape and scale model parameters are jointly computed, while the pose parameters are estimated separately for each modality. RESULTS: The proposed method is successfully applied to nine pairs of preoperative clinical MR volumes and their corresponding US images. The results are assessed using the target registration error (TRE) metric in both MR and US domains. The s [Formula: see text] p [Formula: see text] s model in the proposed joint registration framework results in a mean TRE of 2.62 and 4.20 mm for MR and US images, respectively, on different landmarks. CONCLUSION: The joint framework benefits from the complementary features in both modalities, leading to significantly smaller TREs compared to a model-to-US registration approach. The s [Formula: see text] p [Formula: see text] s model also outperforms our previous shape [Formula: see text] pose model of the lumbar spine, as separating scale from pose allows to better capture pose and guarantees equally-sized vertebrae in both modalities. Furthermore, the simultaneous visualization of the patient-specific models on the MR and US domains makes it possible for clinicians to better evaluate the local registration accuracy.

Joint registration of ultrasound, CT and a shape+pose statistical model of the lumbar spine for guiding anesthesia.

PURPOSE: Facet joint injections and epidural needle insertions are widely used for spine anesthesia. Accurate needle placement is important for effective therapy delivery and avoiding complications arising from damage of soft tissue and nerves. Needle guidance is usually performed by fluoroscopy or palpation, resulting in radiation exposure and multiple needle re-insertions. Several ultrasound (US)-based approaches have been proposed but have not found wide acceptance in clinical routine. This is mainly due to difficulties in interpretation of the complex spinal anatomy in US, which leads to clinicians' lack of confidence in relying only on information derived from US for needle guidance. METHODS: We introduce a multimodal joint registration technique that takes advantage of easy-to-interpret preprocedure computed topography (CT) scans of the lumbar spine to concurrently register a shape+pose model to the intraprocedure 3D US. Common shape coefficients are assumed between two modalities, while pose coefficients are specific to each modality. RESULTS: The joint method was evaluated on patient data consisting of ten pairs of US and CT scans of the lumbar spine. It was successfully applied in all cases and yielded an RMS shape error of 2.1 mm compared to the CT ground truth. The joint registration technique was compared to a previously proposed method of statistical model to US registration Rasoulian et al. (Information processing in computer-assisted interventions. Springer, Berlin, pp 51-60, 2013). The joint framework improved registration accuracy to US in 7 out of 17 visible vertebrae, belonging to four patients. In the remaining cases, the two methods were equally accurate. CONCLUSION: The joint registration allows visualization and augmentation of important anatomy in both the US and CT domain and improves the registration accuracy in both modalities. Observing the patient-specific model in the CT domain allows the clinicians to assess the local registration accuracy qualitatively, which is likely to increase their confidence in using the US model for deriving needle guidance decisions.