Automatic classification framework for ventricular septal defects: a pilot study on high-throughput mouse embryo cardiac phenotyping

Abstract

Intensive international efforts are underway toward phenotyping the entire mouse genome by modifying all its [Formula: see text] genes one-by-one for comparative studies. A workload of this scale has triggered numerous studies harnessing image informatics for the identification of morphological defects. However, existing work in this line primarily rests on abnormality detection via structural volumetrics between wild-type and gene-modified mice, which generally fails when the pathology involves no severe volume changes, such as ventricular septal defects (VSDs) in the heart. Furthermore, in embryo cardiac phenotyping, the lack of relevant work in embryonic heart segmentation, the limited availability of public atlases, and the general requirement of manual labor for the actual phenotype classification after abnormality detection, along with other limitations, have collectively restricted existing practices from meeting the high-throughput demands. This study proposes, to the best of our knowledge, the first fully automatic VSD classification framework in mouse embryo imaging. Our approach leverages a combination of atlas-based segmentation and snake evolution techniques to derive the segmentation of heart ventricles, where VSD classification is achieved by checking whether the left and right ventricles border or overlap with each other. A pilot study has validated our approach at a proof-of-concept level and achieved a classification accuracy of 100% through a series of empirical experiments on a database of 15 images.

Keywords: atlas-based segmentation; mouse embryo phenotyping; snake evolution; ventricular septal defects.

Fig. 1

Ventricular septal defects (VSD) pathology: (a) a negative/healthy subject image and (b) a positive/defective sample. The left and right ventricles are separated in the former case, but are interconnected in the latter case.

Fig. 2

Example raw image of mouse embryo (a) in axial view, (b) in sagittal view, and (c) its intensity histogram.

Fig. 3

Creation of the wild-type mean image through a sequence of group-wise image alignment and averaging processes with rigid, affine, and nonlinear registrations using all the available wild-type images.

Fig. 4

Feature function superimposed with the intensity histogram of heart region: each voxel is projected to the feature space and derives a value based on its intensity with respect to the feature function.

Fig. 5

Example mouse embryo extraction result: (a) a raw image superimposed with the label map by our algorithm, (b) the embryo image after extraction, and (c) three-dimensional reconstruction of the extracted embryo.

Fig. 6

Template updates (a) right after group-wise linear alignment and averaging, (b) after 5 iterations, (c) 8 iterations, and (d) 10 iterations of nonlinear processing.

Fig. 7

(a) A two-dimensional (2-D) image slice of the created heart atlas: the red-colored area represents the heart. (b) A 2-D slice of the created ventricle atlas: the green and red areas, respectively, represent the left and right ventricles.

Fig. 8

Example heart segmentation results of four target images: (a) and (b) two VSD positive samples, and (c) and (d) two VSD negative samples. N.B. only a relatively coarse accuracy is attained in this stage, as heart segmentation merely serves an initialization mask for ventricle segmentation.

Fig. 9

Ventricle segmentation and VSD classification results: (a) and (c) a VSD negative and two positive cases, and (d) to (f) the images superimposed with corresponding label maps. Fluid in left/right ventricle is colored in red/blue accordingly. N.B. in (b), the blood was mostly flushed away during data acquisition, which enables contour/snake to readily expand from one ventricle to the other, resulting in almost entirely overlapping segmentations, whereas in (c), there was a significant amount of blood residuals, leading to a low level of contour/snake overlap after evolution.

Fig. 10

Results of snake evolution on image (a) with the settings (b) $\alpha =0.9$, $\beta =0.1$, $\mathit{max}Itr=500$ and (c) $\alpha =0.6$, $\beta =0.2$, $\mathit{max}Itr=500$. The former successfully identified ventricular connectivity, while the latter did not.