Generalizable Framework for Atrial Volume Estimation for Cardiac CT Images Using Deep Learning With Quality Control Assessment

Abstract

ObjectivesCardiac computed tomography (CCT) is a common pre-operative imaging modality to evaluate pulmonary vein anatomy and left atrial appendage thrombus in patients undergoing catheter ablation (CA) for atrial fibrillation (AF). These images also allow for full volumetric left atrium (LA) measurement for recurrence risk stratification, as larger LA volume (LAV) is associated with higher recurrence rates. Our objective is to apply deep learning (DL) techniques to fully automate the computation of LAV and assess the quality of the computed LAV values.MethodsUsing a dataset of 85,477 CCT images from 337 patients, we proposed a framework that consists of several processes that perform a combination of tasks including the selection of images with LA from all other images using a ResNet50 classification model, the segmentation of images with LA using a UNet image segmentation model, the assessment of the quality of the image segmentation task, the estimation of LAV, and quality control (QC) assessment.ResultsOverall, the proposed LAV estimation framework achieved accuracies of 98% (precision, recall, and F1 score metrics) in the image classification task, 88.5% (mean dice score) in the image segmentation task, 82% (mean dice score) in the segmentation quality prediction task, and R2 (the coefficient of determination) value of 0.968 in the volume estimation task. It correctly identified 9 out of 10 poor LAV estimations from a total of 337 patients as poor-quality estimates.ConclusionsWe proposed a generalizable framework that consists of DL models and computational methods for LAV estimation. The framework provides an efficient and robust strategy for QC assessment of the accuracy for DL-based image segmentation and volume estimation tasks, allowing high-throughput extraction of reproducible LAV measurements to be possible.

Using machine learning to enhance prediction of atrial fibrillation recurrence after catheter ablation

“…LAV was generated using a deep learning framework described in detail elsewhere 13 . In brief, the framework begins by using a ResNet50 classification model to select images within the series that include the left atrium and then uses a UNet image segmentation model to select the left atrium.…”

Predicting post-contrast information from contrast agent free cardiac MRI using machine learning: Challenges and methods

“…For the UNet segmentation model whose input are the cine SAX frames and whose output is the segmentation mask that marks the regions of the epicardium, endocardium and the scar, the overall performance based on the average DSC score for the 108 test set is 0.75 (±0.20) for the endocardium, 0.51 (±0.21) for the epicardium and 0.20 (±0.17) for the scar; and 0.24 (±0.12), 0.48 (±0.23), and 0.77 (±0.18) for the scar, epicardium and endocardium, respectively, from fivefold cross-validation. At first glance, the prediction accuracies of the epicardium or endocardium appear low given that medical image segmentation is a widely studied subject ( 3 ) and the state-of-the-art average DSC could reach 0.94 for CMR ( 52 ) or 0.885 for cardiac CT images ( 53 ). We note that in those previous studies, the models (i.e., the one-input and one-output models) involved a single image and the prediction of the models is compared with the ground truth obtained after segmenting the input image.…”

“…Of particular note here is the precision score ( , i.e., 0.19 – see also the confusion matrix). ResNet is a very successful DL architecture ( 3 , 53 ) and the low performance of our ResNet model in this case (which could be for several reasons, e.g., not enough dataset or not information from the given dataset that will enable the model learn the underlying model parameters) emphasize difficulty and complexity of the problem we intend to solve. The lack of sufficient information could also be as a result of reduction of the resolution of the cine SAX images due to resizing (i.e., 224 × 224) although resizing can sometimes be necessary due to hardware limitations or to ensure all input images have common size.…”

Novel Domain Knowledge Encoding Enables Machine Learning of Rapid, Expert-level Segmentation of Cardiac Computed Tomography

“…There is a growing literature on ML-based segmentation of cardiac CT 8,32,33 , to guide therapy such as ablation for AF or other atrial arrhythmias, but also to predict clinical endpoints such as the risk of AF recurrence 33,34 . However, most studies that segmented the LA body did not specifically segment the PVs and LAA 32,33 .…”