Regional Heart Motion Abnormality Detection via Multiview Fusion

Abstract

Abstract. This study investigates regional heart motion abnormality detection via multiview fusion in cine cardiac MR images. In contrast to previous methods which rely only on short-axis image sequences, the proposed approach exploits the information from several other long-axis image sequences, namely, 2-chamber, 3-chamber and 4-chamber MR images. Our analysis follows the standard issued by American Heart Association to identify 17 standardized left ventricular segments. The proposed method first computes an initial sequence of corresponding myocardial points using a nonrigid image registration algorithm within each sequence. Then, these points were mapped to 3D space and tracked using Unscented Kalman Filter (UKS). We propose a maximum likelihood based track-to-track fusion approach to combine UKS tracks from multiple image views. Finally, we use a Shannon's differential entropy of distributions of potential classifiers obtained from multiview fusion estimates, and a naive Bayes classifier algorithm to automatically detect abnormal functional regions of the myocardium. We proved the benefits of the proposed method by comparing the classification results with and without fusion over 480 regional myocardial segments obtained from 30 subjects. The evaluations in comparisons to the ground truth classifications by radiologists showed that the proposed fusion yielded an area-under-the-curve (AUC) of 95.9%, bringing a significant improvement of 3.8% in comparisons to previous methods that use only short-axis images.

Cardiac Motion Scoring with Segment- and Subject-Level Non-local Modeling

“…Modality Training/Test Accuracy(%) Kappa value [11] 30 SAX MR LOO 90.8 0.738 [9] 30 SAX, 2C, 3C, 4C MR LOO 91.9 0.738 [10] 58 SAX MR LOO 87.1 0.73 [14,13] 22 SAX Tagged MR LOO 87.8 - [6] 17 SAX Basal MR No split 86.3 0.693 [1] 58 SAX MR 3-fold CV 86 0.73 [15] 89 SAX MR 45/44 65.9 - [5] 129 2C and 4C Echo 65/64 76.5 -SAX: short axis view; 2C: two chamber view; 3C: three chamber view; 4C: four chamber view; LOO: leave-one-subject-out; CV: cross validation the presence of different types of pathology. In routine clinical practice, motion scoring is often conducted by labor-intensive visual inspection of the dynamic cardiac sequences of MR or Echocardiograms for each segment of left ventricle, following the scoring system of [4]: (1) normal, (2) hypokinetic, (3) akinetic, and (4) dyskinetic.…”

“…However, automated motion scoring from cardiac MR images has never been investigated despite its clinical significance. Existing work only focused on binary motion abnormality detection [1,2,5,6,7,9,10,11,12,13,14,15], which alleviates a lot the difficulty of differentiating the subtle difference among various motion patterns. In summary, these methods follow a pipeline of: 1) myocardium segments localization, by manually or semi-automatically delineating the contours of myocardium [1,2,5,6,15,9,10,11,12], 2) handcrafted motion information extraction, including spatial-temporal profiles [13,14], inter-frame correlations [2,1,6], or parameter distribution [9,10,11,12], and then 3) motion classification.…”

“…While these methods achieved promising accuracy for abnormality detection, they cannot be applied to the task of motion scoring because 1) only heuristic handcrafted feature is not sufficient to capture the complex regional motion information; and 2) absence of dependency modeling between local features of all segments cannot capture the subtle motion difference between segments. Besides, they still suffer from the following limitations: 1) delineation of myocardium border is required, either manually or semi-automatically, introducing inconvenience in clinical practice; 2) multiple classifiers are trained for different segments [5,6,2,1] or slices [9,10,11,15], complicating their practical application; 3) only achieves good performance on small dataset (as shown in Table . 1), where the diversity of cardiac motion pattern is limited.…”

Comparative Study of 2D-Cine and 3D-wh Volumetry: Revealing Systemic Error of 2D-Cine Volumetry

“…In contrast, the 3D-wh method does not rely on interpolation; the data capture of the entire heart occurs in one measurement, each at a precisely defined trigger time, in ED or ES [32]. Here, too, movement artifacts negatively affect image acquisition, but artifacts lead to a deterioration of image quality and can be measured by SNR and SN.…”