Convolutional neural network regression for short-axis left ventricle segmentation in cardiac cine MR sequences

Fully automated segmentation of the left ventricle in cine cardiac MRI using neural network regression

“…In particular, apical slices are very challenging due to the relatively small size of the blood pool (including disappearance during systolic phases). We did not find a statistically significant difference between the previous and current work in the MHD metric for apical slices ( P = 0.027). However, we suspect this is because the LVSC consensus validation dataset (CS*) does not include ground truth data for many apical slices.…”

“…This is particularly notable for the MB network, where the myocardial boundaries are delineated as individual radial points inferred from a polar transform of the input image centered on the LV centroid, as opposed to the more common technique of myocardial segmentation by per‐pixel classification. In our previous work, we found this regression technique superior to other state‐of‐the‐art per‐pixel classification networks; it implicitly enforces useful physiological constraints in the model, such as there being only a single connected object, and that the endo‐ and epicardium contours share a common centerpoint …”

“…Our system primarily operates on 2D intensity images, as well as 2D first harmonic magnitude images, H1 mag , obtained by applying a 1D Fourier transform across the temporal dimension of a 2D+time slice. In our previous work we found H1 mag to efficiently incorporate temporal information for SA volumes …”

“…Our earlier work utilized two networks (CTR and MB) with differing architecture . In particular, the earlier MB network utilized a specific "coarse and fine" dual subnetwork design with varying input windows; the total number of parameters for both networks was around three million.…”

Unsupervised Myocardial Segmentation for Cardiac BOLD

“…Finally briefly for completeness we mention deep learning methods that are fully supervised and aim to extract a hierarchy of image features at multiple scales (e.g. see [30], [31], [32], [33], [34], and a recent review [35]).…”