automatic detection of aortic valve pathology using 3-chamber cine MRI

Abstract Funding Acknowledgements Type of funding sources: None. Introduction The assessment aortic valve pathology using magnetic resonance imaging (MRI) typically relies on blood velocity estimates acquired phase contrast (PC) MRI. However, abnormalities in flow through the often manifest by dephasing signal gated balanced steady-state free precession (bSSFP) scans (cine MRI) [1]. This suggests cine MRI may be used to identify patients with who would benefit from additional detailed PC-MRI scans. Nevertheless, there are currently no tools automatically root which can improve diagnosis and optimise clinical CMR protocols. Aim To determine whether a classification neural network (NN) detect 3-chamber MRI, without need for dedicated or manual annotation. Data We train test our approach retrospective dataset three UK hospitals, single-slice bSSFP (cardiac frames: 31 ± 15, spatial resolution: 1.17–1.56 x 8 mm3) n = 577 patients. Using information assessments, each image was labelled as: regurgitation (AR) (13%), stenosis (AS) (11%), mixed disease (MVD) (9%), showing relevant (67%). 322 randomly selected images were training, 81 validation 174 testing. Methods trained 3D NN (DenseNet201 model [2] 4 dense blocks include (i.e., spatiotemporal) convolutional layers; Adam optimisation cross-entropy loss function) classify time-resolved MR images. binary labels: "pathology" (corresponding AR, AS MVD) "no pathology". As pre-processing step, we identified heart all frames an in-house algorithm that detected motion across cardiac cycle, cropped remove non-cardiac anatomical structures. After classification, gradient-weighted class activation mapping (Grad-CAM) [3] gain confidence most dependent voxels close root. Fig 1 shows overview processing pipeline. Results Our accurately predicts MRI: 2 confusion matrix evaluation metrics study. Grad-CAM heatmaps showed region contributed as seen 1. Conclusion Convolutional NNs annotations. protocols, namely triage tool likely sequences.