Echocardiographic artificial intelligence for pulmonary hypertension classification.

OBJECTIVE: The classification of pulmonary hypertension (PH) is crucial for determining the appropriate therapeutic strategy. We investigated whether machine learning (ML) algorithms may assist in echocardiographic PH prediction, where current guidelines recommend integrating several different parameters. METHODS: We obtained physical and echocardiographic data from 885 patients who underwent right heart catheterisation (RHC). Patients were classified into three groups: non-PH, precapillary PH and postcapillary PH, based on values obtained from RHC. Using 24 parameters, we created predictive models employing four different classifiers and selected the one with the highest area under the curve. We then calculated the macro-average classification accuracy for PH on the derivation cohort (n=720) and prospective validation data set (n=165), comparing the results with guideline-based echocardiographic assessment obtained from each cohort. RESULTS: Logistic regression with elastic net regularisation had the highest classification accuracy, with areas under the curves of 0.789, 0.766 and 0.742 for normal, precapillary PH and postcapillary PH, respectively. The ML model demonstrated significantly better predictive accuracy than the guideline-based echocardiographic assessment in the derivation cohort (59.4% vs 51.6%, p<0.01). In the independent validation data set, the ML model's accuracy was comparable to the guideline-based PH classification (59.4% vs 57.8%, p=0.638). CONCLUSIONS: This preliminary study suggests promising potential for our ML model in predicting echocardiographic PH. Further research and validation are needed to fully assess its clinical utility in PH diagnosis and treatment decision-making.

Unsupervised cluster analysis reveals different phenotypes in patients after transcatheter aortic valve replacement.

Aims: The aim of this study was to identify phenotypes with potential prognostic significance in aortic stenosis (AS) patients after transcatheter aortic valve replacement (TAVR) through a clustering approach. Methods and results: This multi-centre retrospective study included 1365 patients with severe AS who underwent TAVR between January 2015 and March 2019. Among demographics, laboratory, and echocardiography parameters, 20 variables were selected through dimension reduction and used for unsupervised clustering. Phenotypes and outcomes were compared between clusters. Patients were randomly divided into a derivation cohort (n = 1092: 80%) and a validation cohort (n = 273: 20%). Three clusters with markedly different features were identified. Cluster 1 was associated predominantly with elderly age, a high aortic valve gradient, and left ventricular (LV) hypertrophy; Cluster 2 consisted of preserved LV ejection fraction, larger aortic valve area, and high blood pressure; and Cluster 3 demonstrated tachycardia and low flow/low gradient AS. Adverse outcomes differed significantly among clusters during a median of 2.2 years of follow-up (P < 0.001). After adjustment for clinical and echocardiographic data in a Cox proportional hazards model, Cluster 3 (hazard ratio, 4.18; 95% confidence interval, 1.76-9.94; P = 0.001) was associated with increased risk of adverse outcomes. In sequential Cox models, a model based on clinical data and echocardiographic variables (χ2 = 18.4) was improved by Cluster 3 (χ2 = 31.5; P = 0.001) in the validation cohort. Conclusion: Unsupervised cluster analysis of patients after TAVR revealed three different groups for assessment of prognosis. This provides a new perspective in the categorization of patients after TAVR that considers comorbidities and extravalvular cardiac dysfunction.

Deep learning approach for analyzing chest x-rays to predict cardiac events in heart failure.

Background: A deep learning (DL) model based on a chest x-ray was reported to predict elevated pulmonary artery wedge pressure (PAWP) as heart failure (HF). Objectives: The aim of this study was to (1) investigate the role of probability of elevated PAWP for the prediction of clinical outcomes in association with other parameters, and (2) to evaluate whether probability of elevated PAWP based on DL added prognostic information to other conventional clinical prognostic factors in HF. Methods: We evaluated 192 patients hospitalized with HF. We used a previously developed AI model to predict HF and calculated probability of elevated PAWP. Readmission following HF and cardiac mortality were the primary endpoints. Results: Probability of elevated PAWP was associated with diastolic function by echocardiography. During a median follow-up period of 58 months, 57 individuals either died or were readmitted. Probability of elevated PAWP appeared to be associated with worse clinical outcomes. After adjustment for readmission score and laboratory data in a Cox proportional-hazards model, probability of elevated PAWP at pre-discharge was associated with event free survival, independent of elevated left atrial pressure (LAP) based on echocardiographic guidelines (p < 0.001). In sequential Cox models, a model based on clinical data was improved by elevated LAP (p = 0.005), and increased further by probability of elevated PAWP (p < 0.001). In contrast, the addition of pulmonary congestion interpreted by a doctor did not statistically improve the ability of a model containing clinical variables (compared p = 0.086). Conclusions: This study showed the potential of using a DL model on a chest x-ray to predict PAWP and its ability to add prognostic information to other conventional clinical prognostic factors in HF. The results may help to enhance the accuracy of prediction models used to evaluate the risk of clinical outcomes in HF, potentially resulting in more informed clinical decision-making and better care for patients.

Classification of chest X-ray images by incorporation of medical domain knowledge into operation branch networks.

BACKGROUND: This study was conducted to alleviate a common difficulty in chest X-ray image diagnosis: The attention region in a convolutional neural network (CNN) does not often match the doctor's point of focus. The method presented herein, which guides the area of attention in CNN to a medically plausible region, can thereby improve diagnostic capabilities. METHODS: The model is based on an attention branch network, which has excellent interpretability of the classification model. This model has an additional new operation branch that guides the attention region to the lung field and heart in chest X-ray images. We also used three chest X-ray image datasets (Teikyo, Tokushima, and ChestX-ray14) to evaluate the CNN attention area of interest in these fields. Additionally, after devising a quantitative method of evaluating improvement of a CNN's region of interest, we applied it to evaluation of the proposed model. RESULTS: Operation branch networks maintain or improve the area under the curve to a greater degree than conventional CNNs do. Furthermore, the network better emphasizes reasonable anatomical parts in chest X-ray images. CONCLUSIONS: The proposed network better emphasizes the reasonable anatomical parts in chest X-ray images. This method can enhance capabilities for image interpretation based on judgment.

Deep Learning for Detection of Exercise-Induced Pulmonary Hypertension Using Chest X-Ray Images.

Background: Stress echocardiography is an emerging tool used to detect exercise-induced pulmonary hypertension (EIPH). However, facilities that can perform stress echocardiography are limited by issues such as cost and equipment. Objective: We evaluated the usefulness of a deep learning (DL) approach based on a chest X-ray (CXR) to predict EIPH in 6-min walk stress echocardiography. Methods: The study enrolled 142 patients with scleroderma or mixed connective tissue disease with scleroderma features who performed a 6-min walk stress echocardiographic test. EIPH was defined by abnormal cardiac output (CO) responses that involved an increase in mean pulmonary artery pressure (mPAP). We used the previously developed AI model to predict PH and calculated PH probability in this cohort. Results: EIPH defined as ΔmPAP/ΔCO >3.3 and exercise mPAP >25 mmHg was observed in 52 patients, while non-EIPH was observed in 90 patients. The patients with EIPH had a higher mPAP at rest than those without EIPH. The probability of PH based on the DL model was significantly higher in patients with EIPH than in those without EIPH. Multivariate analysis showed that gender, mean PAP at rest, and the probability of PH based on the DL model were independent predictors of EIPH. A model based on baseline parameters (age, gender, and mPAP at rest) was improved by adding the probability of PH predicted by the DL model (AUC: from 0.65 to 0.74; p = 0.046). Conclusion: Applying the DL model based on a CXR may have a potential for detection of EIPH in the clinical setting.

Deep Learning for Detection of Elevated Pulmonary Artery Wedge Pressure Using Standard Chest X-Ray.

BACKGROUND: To accurately diagnose and control heart failure (HF), it is important to carry out a simple assessment of elevated pulmonary arterial wedge pressure (PAWP). The aim of this study was to develop and validate an objective method for detecting elevated PAWP by applying deep learning (DL) to a chest x-ray (CXR). METHODS: We enrolled 1013 consecutive patients with a right-heart catheter between October 2009 and February 2020. We developed a convolutional neural network to identify patients with elevated PAWP (> 18 mm Hg) as the actual value of PAWP to be used in the dataset for training. In the prospective validation dataset used to detect elevated PAWP, the area under the receiver operating characteristic curve (AUC) was calculated using the DL model that evaluated the CXR. RESULTS: In the prospective validation dataset, the AUC of the DL model with CXR was not significantly different from the AUC produced by brain natriuretic peptide (BNP) and the echocardiographic left-ventricular diastolic dysfunction (DD) algorithm (DL model: 0.77 vs BNP: 0.77 vs DD algorithm: 0.70; respectively; P = NS for all comparisons); it was, however, significantly higher than the AUC of the cardiothoracic ratio (DL model vs cardiothoracic ratio [CTR]: 0.66, P = 0.044). The model based on 3 parameters (BNP, DD algorithm, and CTR) was improved by adding the DL model (AUC: from 0.80 to 0.86; P = 0.041). CONCLUSIONS: Applying the DL model based on a CXR (a classical, universal, and low-cost test) is useful for screening for elevated PAWP.

Deep learning to predict elevated pulmonary artery pressure in patients with suspected pulmonary hypertension using standard chest X ray.

Accurate diagnosis of pulmonary hypertension (PH) is crucial to ensure that patients receive timely treatment. We hypothesized that application of artificial intelligence (AI) to the chest X-ray (CXR) could identify elevated pulmonary artery pressure (PAP) and stratify the risk of heart failure hospitalization with PH. We retrospectively enrolled a total of 900 consecutive patients with suspected PH. We trained a convolutional neural network to identify patients with elevated PAP (> 20 mmHg) as the actual value of PAP. The endpoints in this study were admission or occurrence of heart failure with elevated PAP. In an independent evaluation set for detection of elevated PAP, the area under curve (AUC) by the AI algorithm was significantly higher than the AUC by measurements of CXR images and human observers (0.71 vs. 0.60 and vs. 0.63, all p < 0.05). In patients with AI predicted PH had 2-times the risk of heart failure with PH compared with those without AI predicted PH. This preliminary work suggests that applying AI to the CXR in high risk groups has limited performance when used alone in identifying elevated PAP. We believe that this report can serve as an impetus for a future large study.

Classification of optical coherence tomography images using a capsule network.

BACKGROUND: Classification of optical coherence tomography (OCT) images can be achieved with high accuracy using classical convolution neural networks (CNN), a commonly used deep learning network for computer-aided diagnosis. Classical CNN has often been criticized for suppressing positional relations in a pooling layer. Therefore, because capsule networks can learn positional information from images, we attempted application of a capsule network to OCT images to overcome that shortcoming. This study is our attempt to improve classification accuracy by replacing CNN with a capsule network. METHODS: From an OCT dataset, we produced a training dataset of 83,484 images and a test dataset of 1000 images. For training, the dataset comprises 37,205 images with choroidal neovascularization (CNV), 11,348 with diabetic macular edema (DME), 8616 with drusen, and 26,315 normal images. The test dataset has 250 images from each category. The proposed model was constructed based on a capsule network for improving classification accuracy. It was trained using the training dataset. Subsequently, the test dataset was used to evaluate the trained model. RESULTS: Classification of OCT images using our method achieved accuracy of 99.6%, which is 3.2 percentage points higher than that of other methods described in the literature. CONCLUSION: The proposed method achieved classification accuracy results equivalent to those reported for other methods for CNV, DME, drusen, and normal images.