Fully automatic segmentation of 4D MRI for cardiac functional measurements

Abstract

Purpose: Segmentation of cardiac medical images, an important step in measuring cardiac function, is usually performed either manually or semi-automatically. Fully automatic segmentation of the left ventricle (LV), the right ventricle (RV) as well as the myocardium of 3D magnetic resonance (MR) images throughout the entire cardiac cycle (4D), remains challenging. This study proposes a deformable-based segmentation methodology for efficiently segmenting 4D (3D+t) cardiac MR images. Methods: The proposed methodology first used the Hough transform and the local Gaussian distribution method (LGD) to segment the LV endocardial contours from cardiac MR images. Following that a novel level set-based shape prior method was applied to generate the LV epicardial contours and the RV boundary. Results: This automatic image segmentation approach has been applied to studies on 17 subjects. The results demonstrated that the proposed method was efficient compared to manual segmentation, achieving a segmentation accuracy with average Dice values of 88.62±5.47%, 87.35±7.26%, and 82.63±6.22% for the LV endocardial, LV epicardial and RV contours, respectively. Conclusions: We have presented a method for accurate LV and RV segmentation. Compared to three existing methods, the proposed method can successfully segment the LV and yields the highest Dice value. This makes it an option for clinical assessment of the volume, size and thickness of the ventricles.

Diagnosis of Left Ventricular Diastolic Dysfunction Using Cardiac Magnetic Resonance Imaging: Comparison of Volume-Time Curves Derived from Long- and Short-Axis Cine Steady-State Free Precession Datasets

“…SAX cine steady-state free precession (SSFP) images are, on the other hand, acquired by default, but generation of volume-time curves relies on assessment of all images across the cardiac cycle. Although substantial progress in semi-automatic segmentation has been made over the last decades, current practice for cine-analysis still comprises substantial input in manual contour tracing or starting and controlling semiautomatic evaluation programs [7][8][9]. This procedure can become very time-consuming, if manual assessment of the LV endocardial contour in all SAX timeframes is required.…”

Left ventricular geometry during unloading and the end-systolic pressure volume relationship: Measurement with a modified real-time MRI-based method in normal sheep

“…LSSPM was designed for cardiac segmentation and functional measurements by Wang et al [19]. In our case, manual segmentation was performed in a blinded manner by 2 readers who contoured 200 slices from all 10 subjects (20 slices for each case) as shown in Fig 3A. A second set of manual contours was drawn by each of the readers in 3 of the cases allowing the calculation of both inter-and intra-observer variability.…”

“…In the current study, an LSSPM software tool proposed by Wang et al [19] was used to automatically segment the LV during VCO. As above, the LSSPM has been validated with an MRXCAT phantom [25] and 17 human MRI studies with over 5000 2D images and achieved the highest Dice value [26] compared with three state-of-the-art published methods.…”

“…Both active contour methods [19][20][21] and deep learning methods [22][23][24] have been successfully used for automatic segmentation of the LV cavity and wall. Deep learning methods require training sets that are large and rich while active contour methods which are usually based on image intensity and shape do not require training sets.…”

“…Deep learning methods require training sets that are large and rich while active contour methods which are usually based on image intensity and shape do not require training sets. Given that training sets of the LV cavity during VCO would be difficult to obtain, in the current study we chose to employ the active contour-type level set shape prior method (LSSPM) described by Wang et al [19] for automatic segmentation of the LV during VCO. LSSPM has been validated with an MRXCAT phantom [25] and 17 human MRI studies with over 5000 2D images and achieved the highest Dice value [26] compared with three state-of-the-art published methods.…”