Multi-Channel Volumetric Neural Network for Knee Cartilage Segmentation in Cone-beam CT
Authors:
Jennifer Maier,
Luis Carlos Rivera Monroy,
Christopher Syben,
Ye** Jeon,
Jang-Hwan Choi,
Mary Elizabeth Hall,
Marc Levenston,
Garry Gold,
Rebecca Fahrig,
Andreas Maier
Abstract:
Analyzing knee cartilage thickness and strain under load can help to further the understanding of the effects of diseases like Osteoarthritis. A precise segmentation of the cartilage is a necessary prerequisite for this analysis. This segmentation task has mainly been addressed in Magnetic Resonance Imaging, and was rarely investigated on contrast-enhanced Computed Tomography, where contrast agent…
▽ More
Analyzing knee cartilage thickness and strain under load can help to further the understanding of the effects of diseases like Osteoarthritis. A precise segmentation of the cartilage is a necessary prerequisite for this analysis. This segmentation task has mainly been addressed in Magnetic Resonance Imaging, and was rarely investigated on contrast-enhanced Computed Tomography, where contrast agent visualizes the border between femoral and tibial cartilage. To overcome the main drawback of manual segmentation, namely its high time investment, we propose to use a 3D Convolutional Neural Network for this task. The presented architecture consists of a V-Net with SeLu activation, and a Tversky loss function. Due to the high imbalance between very few cartilage pixels and many background pixels, a high false positive rate is to be expected. To reduce this rate, the two largest segmented point clouds are extracted using a connected component analysis, since they most likely represent the medial and lateral tibial cartilage surfaces. The resulting segmentations are compared to manual segmentations, and achieve on average a recall of 0.69, which confirms the feasibility of this approach.
△ Less
Submitted 3 December, 2019;
originally announced December 2019.
A Wavelet Approach for the Estimation of Left Ventricular Early Filling Wave Propagation Velocity from Color-M-Mode Echocardiograms
Authors:
Sreyashi Chakraborty,
Hiroyuki Iwano,
Michael E. Hall,
Pavlos P. Vlachos
Abstract:
Objective: This work evaluates a new approach for calculating the left-ventricular (LV) early filling propagation velocity (VP) from color M-Mode (CMM) echocardiograms using wavelet analysis. Unlike current approaches, the method requires no assumptions, user inputs, or heuristic conventions. Background: Current methods for measuring VP using CMM echocardiography do not account for the spatiotempo…
▽ More
Objective: This work evaluates a new approach for calculating the left-ventricular (LV) early filling propagation velocity (VP) from color M-Mode (CMM) echocardiograms using wavelet analysis. Unlike current approaches, the method requires no assumptions, user inputs, or heuristic conventions. Background: Current methods for measuring VP using CMM echocardiography do not account for the spatiotemporal variation of the filling wave propagation velocity. They are instead confined by empirical assumptions and user inputs that significantly hinder the accuracy of VP, subsequently limiting its clinical utility. Methods: We evaluated three methods for measuring LV early filling VP: conventional VP, the strength of propagation (VS), and VP determined from the most energetically important wave (Peak-VW), using 125 patients (Group A) with normal filling (n=50), impaired relaxation (n=25), pseudonormal filling (n=25) and restrictive filling (n=25), and in 69 patients (Group B) with normal (n=32), dilated (n=15), and hypertrophic ventricles (n=22). Results: Peak-VW most accurately distinguished normal left ventricles from diseased ones. Specifically, for Group A which spanned the stages of diastolic function, using receiver operator characteristics (ROC) and measuring their corresponding area under the curve (AUC), the AUC for Peak-VW was 0.92, versus 0.62 for conventional VP, 0.63 for VS and 0.58 for intraventricular pressure difference (IVPD). These correspond to a 50-70% improvement in classification ability. Similar improvements were found using Group B. Conclusion: A new determination of the LV early filling using wavelet analysis, Peak-VW, may provide a more accurate evaluation of diastolic function than the standard method of determining Vp and enable better diagnostic classification of patients suffering with diastolic dysfunction.
△ Less
Submitted 25 October, 2019;
originally announced October 2019.
