Embryonic heart morphogenesis (EHM) is a complex and dynamic process where the heart transforms from a single tube into a four-chambered pump. This process is of great biological and clinical interest but is still poorly understood for two main reasons. On the one hand, the existing imaging modalities for investigating EHM suffered from either limited penetration depth or limited spatial resolution. On the other hand, current works typically adopted manual segmentation, which was tedious, subjective, and time consuming considering the complexity of developing heart geometry and the large size of images. In this paper, we propose to utilize confocal microscopy imaging with tissue optical immersion clearing technique to image the heart at different stages of development for EHM study. The imaging method is able to produce high spatial resolution images and achieve large penetration depth at the same time. Furthermore, we propose a novel convex active contour model for automatic image segmentation. The model has the ability to deal with intensity fall-off in depth which is characterized by confocal microscopy images. We acquired the images of embryonic quail hearts from day 6 to day 14 of incubation for EHM study. The experimental results were promising and provided us with an insight view of early heart growth pattern and also paved the road for data-driven heart growth modeling.
The heart is the first functioning organ in the embryo. Although the morphology of the heart changes dramatically during development where it transforms from a single tube into a four-chambered pump, the heart functions without interruption to serve the metabolic needs of the rapidly growing embryo [
Despite a large body of research in the last decades [
In view of the problems, we propose a new imaging approach for studying EHM, utilizing tissue optical immersion clearing and 3D confocal microscopy imaging, which can produce high spatial resolution images and achieve large penetration depth at the same time. Furthermore, considering the intensity fall-off in depth nature of confocal microscopy images, we propose a convex active contour model with image depth information for automatic image segmentation. A recently proposed Split Bregman method was used to minimize the objective function of the model [
The human heart becomes a four-chambered organ by approximately week 8, which is almost the same time the embryo can be visualized through ultrasound, a point that is too late to visualize EHM [
Optical imaging method has been widely used as a tool for clinical functional imaging owing to its unique informative features, simplicity, safety, and low cost compared to conventional X-ray, MRI, and ultrasound imaging. However, the main limitations of optical imaging techniques, including confocal microscopy, are low contrast and spatial resolution, as well as a small probing depth due to strong light scattering in tissue layers [
In this paper, all the experiments conformed to the
High spatial resolution images were obtained after heart scanning, which had an intraslice pixel size of 1.75
3D view of embryonic quail hearts. From left to right: day 6, day 8, and day 14.
In a confocal microscope, a pinhole is used to reject most out-of-focus light. Thus, the amount of light reaching the detector is low, and the noise statistics can be well described by a Poisson process [
The purpose of image segmentation is to find a partition
The first term on the right side of (
There are total three unknown variables in our model:
To minimize (
To minimize (
To solve the constrained problem in (
Here,
By using Gauss-Seidel iterative scheme, we can get an approximate solution for a 3D variable
After calculating an approximate
Once
As a summary, the procedures of using Split Bregman method to solve ( Initialization: Fix Update Update Update Convergence test: test whether a stable solution The objects are detected by thresholding
In this study, we selected three groups of quail hearts. Each group had five embryonic quail hearts at the development stage of days 6, 7, 8, 9, and 14, respectively. All the hearts were processed and imaged according to Section
Figure
Visual comparison between manual segmentation and automatic segmentation. (a) Original 3D image and three slices in different views. (b) Manual segmentation done by the first biologist. (c) Manual segmentation done by the second biologist. (d) Automatic segmentation.
We also quantitatively evaluate our algorithm by measuring the overlap of automatic segmentation and manual segmentation by using Dice’s similarity coefficient (DSC). For two segmentations
DSC values that measure the overlap between the two manual segmentations, the first manual segmentation against automatic segmentation and the second manual segmentation against automatic segmentation.
Day 6 | Day 7 | Day 8 | Day 9 | Day 14 | |
---|---|---|---|---|---|
Biologist 1 versus biologist 2 | 0.75 | 0.79 | 0.85 | 0.87 | 0.93 |
Automatic versus biologist 1 | 0.65 | 0.72 | 0.78 | 0.85 | 0.88 |
Automatic versus biologist 2 | 0.68 | 0.75 | 0.79 | 0.81 | 0.91 |
3D segmentation of one group of the hearts. Columns from left to right are the heart at days 6, 7, 8, 9, and 14. For visualization purpose, the outer boundary is rendered as transparent (L: left ventricle. R: right ventricle).
Figure
We also quantify the average volume of the whole heart and the luminal space at different stages of development based on the segmentation results. We use the open source VTK library (
Average volume of the whole heart and the luminal space at different stages of development (mm3).
Day 6 | Day 7 | Day 8 | Day 9 | Day 14 | |
---|---|---|---|---|---|
Total heart | 2.6 | 3.4 | 6.0 | 10.2 | 77.5 |
Luminal space | 0.41 | 0.62 | 0.75 | 1.74 | 20.2 |
The technique outlined in this paper provided the framework of imaging and automatic segmentation of developing hearts for EHM study. By combining confocal microscopy imaging with optical clearing, our method was able to achieve penetration depth over 6 mm that enabled us to acquire volumetric images of the developing heart through the whole incubation period. We believe this imaging data can help biologists to understand more details of early heart development and investigate events that lead to congenital heart defects.
Image segmentation is always a headache for researchers in this field because of the complexity of the developing geometry. The convex active contour model proposed in this paper was a first step towards automatic segmentation in EHM study and showed promising results. One significant challenge in developing heart segmentation is the lack of a gold standard. Due to the expensive labor cost to label the images, we provided limited validation in the paper. In the future, we will build a larger manually segmented database for segmentation algorithm validation. What is more, this database could also be used for training and testing parameter-free machine learning algorithms.
The ultimate goal of this work will be heart growth modeling. Due to the complexity of developing heart, current heart growth modeling mainly focuses on very early stages of EHM [
We proposed an imaging approach and a novel automatic segmentation method for EHM study. We demonstrated the applicability of our imaging method to capture the 3D structure of embryonic quail hearts and also proved the efficiency of our segmentation algorithm for EHM study in both visual inspection and quantitative analysis. Based on the findings from EHM study, we believe this work could help us to further understand the fundamental mechanisms of embryonic heart development.