Trophectoderm segmentation in human embryo images via inceptioned U-Net
Graphical abstract
Introduction
In the last four decades, the use of fertility services has increased dramatically. Couples are under pressure from delayed childbearing in combination with the natural decline in female reproductive capacity. By the age of 40, there is a significant reduction in women’s fertility and a high risk of miscarriage. The Canadian Community Health Survey (data for ) reports that about 15% (1.8 million) of Canadian women suffer from impaired fecundity. Despite the importance of infertility treatment, in particular for women, there has been little attention to the development of computer-based technologies to improve the treatment and/or outcomes.
In-Vitro Fertilization (IVF) procedure is one of the most common procedures for treating infertility problem. During IVF, a female’s ovaries are hyper-stimulated to produce multiple egg cells for external fertilization. The fertilized eggs (embryos) are incubated in controlled environmental conditions and imaged digitally using Hoffman Modulation Contrast (HMC) microscopes for about 5 to 6 days. These embryos are visually evaluated so that only embryos with the highest potentials are transferred to the uterus.
Clearly, the gold standard/ground truth for embryo quality is a healthy live birth. Quality assessment of an embryo is to identify its potentials leading to a healthy baby. As one of the most effective and commonly used quality assessment methods, morphological evaluation of an embryo is to assign grades for the quality of an embryo based on intrinsic morphological structures of its components such as the density, size and expansion level of various cells.
Another method for assessing an embryo’s quality is preimplantation genetic screening (PGS), which has excellent ability to predict non-implanting embryos (negative predictive value = 96%) (Scott et al., 2012) but modest positive predictive value ( live birth rates) (Scott, Ferry, Su, Tao, Scott, Treff, 2012, Werner, Leondires, Schoolcraft, Miller, Copperman, Robins, et al., 2014). However, utilization of this technology remains low due to incremental cost with embryo biopsy and genetic testing (only 8.2% in Canada in 2015). Although the failure of implantation is often due to whole chromosome aneuploidy, the benefits of Preimplantation Genetic Testing for Aneuploidy (PGT-A) over conventional embryo morphology remains an ongoing debate. While earlier, smaller studies have demonstrated improved embryo implantation rates with PGT-A over morphology (Scott, Upham, Forman, Hong, Scott, Taylor, Tao, Treff, 2013, Yang, Liu, Collins, Salem, Liu, Lyle, Peck, Sills, Salem, 2012), increased costs with this technology and the invasiveness of the biopsy make it a less than an optimal method of embryo selection. More recent and larger studies have demonstrated a lack of benefit with PGT-A over embryo morphology, with no evidence of improved ongoing pregnancy or reduced miscarriage (Zegers-Hochschild et al., 2017b). Therefore, morphology assessment still remains the most common method of embryo/blastocyst selection (Gardner, Stevens, Sheehan, Schoolcraft, 2007, Ahlstrom, Westin, Reismer, Wikland, Hardarson, 2011, Luke, Brown, Stern, Jindal, Racowsky, Ball, 2014).
Here, we propose an automatic approach for segmentation of trophectoderm (TE) region in HMC microscopic day human embryo (also known as a blastocyst) images using a novel deep learning approach. Automatic segmentation of TE regions in a blastocyst is the first step towards automatic grading of the quality of TE. The size and the quality of the TE region are strong indicators of an embryo’s quality (Gardner et al., 2000). The proposed methods outperform the best-reported results in this field to this date.
Before summarizing the previous work for segmentation of TE region, we review some of the specialist terminologies for an embryo’s progress during incubation. According to the 2017 International Glossary on Infertility and Fertility care definitions (Zegers-Hochschild et al., 2017a), an embryo is a biological organism resulting from the development of the zygote, until eight completed weeks after fertilization, equivalent to 10 weeks of gestational age. The blastocyst is the stage of pre-implantation embryo development that occurs around day after insemination. It contains a thick membrane that surrounds the developing embryo (zona pellucida, ZP), an outer layer of cells (trophectoderm, TE), a fluid-filled central cavity (blastocoele), and an inner group of cells (inner cell mass, ICM). ZP protects the embryo during the development and regulates communications of the inside TE (the component that develops into the placenta) and ICM (the component that gives rise to the structures of the fetus). The fluid-filled cavity of the blastocyst, viz., regions between the TE and ICM, is called the blastocoel or cavity. Fig. 1 depicts a sample blastocyst image with highlighted ZP (Fig. 1-a), TE (Fig. 1-b), and ICM (Fig. 1-c) regions highlighted in blue.
Section snippets
Related work
Previous work for automatic segmentation of TE region is limited (Saeedi et al., 2017). The initial attempts for analyzing human blastocysts were mostly semiautomatic (Filho, Noble, Wells, 2010a, Filho, Noble, Poli, Griffiths, Emerson, Wells, 2012). Giusti et al. (2010) introduced a system using variational level-set algorithm (Li et al., 2005) to segment the TE’s inner boundaries. Later, Filho et al. (2012) attempted segmentation of the TE and ICM using level sets (Filho et al., 2010b). Their
Methodology
In semantic segmentation, availability of some knowledge on the global context is beneficial, if not necessary (Luo, Li, Urtasun, Zemel, 2017, Zhang, Liu, Wang, 2017, Li, Liu, Yang, Sun, Hu, Zhang, Li, 2017). Two common ways of increasing the receptive field in a convolutional network are stacking more convolutional layers and adding more down-sampling layers. The former strategy expands the receptive field linearly and the later one expands the receptive field multiplicatively (Luo et al.,
Generating synthetic blastocyst images
In this section, the effect of the size of the training set on the performance of the system is studied and summarized in Fig. 7. Data augmentation and patch-based processing are considered as two conventional ways of increasing the training size. Specifically, augmentation is performed using several randomized processes from the following list:
- •
Flipping horizontally or vertically
- •
Rotating by an angle [0, 360]
- •
Zooming in or out [0.8, 1.2]
- •
Shear transform [0.8, 1.2]
- •
Adding light Gaussian [
Experiment setup
Here, we describe the human blastocyst dataset along with the information on implementation details of the proposed models.
Conclusion
In this paper, four novel models based on deep learning architectures were proposed for TE segmentation. These models introduced four effective ways to increase the receptive field by approximately 60% without losing spatial information while adding only 4% more parameters when compared to the baseline U-Net model. The introduced multi-scaled ensembling technique was able to further improve the performance. Quality synthetic blastocyst images were generated to facilitate better training in the
Declaration of Competing Interest
None.
References (47)
- et al.
Blastomere segmentation and 3d morphology measurements of early embryos from hoffman modulation contrast image stacks
2010 IEEE International Symposium on Biomedical Imaging: From Nano to Macro
(2010) - et al.
Human blastocyst segmentation using neural network
2016 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE)
(2016) - et al.
Adam: A method for stochastic optimization
CoRR
(2014) - et al.
Understanding the effective receptive field in deep convolutional neural networks
CoRR
(2017) - et al.
Automatic identification of human blastocyst components via texture
IEEE Trans. Biomed. Eng.
(2017) - et al.
Blastocyst biopsy with comprehensive chromosome screening and fresh embryo transfer significantly increases in vitro fertilization implantation and delivery rates: a randomized controlled trial
Fertility Sterility
(2013) - et al.
Automatic segmentation of trophectoderm in microscopic images of human blastocysts
IEEE Trans. Biomed. Eng.
(2015) - et al.
Selection of single blastocysts for fresh transfer via standard morphology assessment alone and with array cgh for good prognosis ivf patients: results from a randomized pilot study
Mole. Cytogenet.
(2012) - et al.
The international glossary on infertility and fertility care, 2017
Fertility Sterility
(2017) - et al.
Can signal-to-noise ratio perform as a baseline indicator for medical image quality assessment
IEEE Access
(2018)
Trophectoderm morphology: an important parameter for predicting live birth after single blastocyst transfer
Hum. Reprod.
Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs
IEEE Trans. Pattern Anal. Mach.Intell.
End-to-end adversarial retinal image synthesis
IEEE Transactions on Medical Imaging
A method for semi-automatic grading of human blastocyst microscope images.
Hum. Reprod.
A review on automatic analysis of human embryo microscope images
Open Biomed. Eng. J.
Toward a method for automatic grading of microscope human embryo images
2010 IEEE International Symposium on Biomedical Imaging: From Nano to Macro
Analysis of blastocyst morphology
Hum. Preimplant. Embryo Sel.
Selection of single blastocysts for fresh transfer via standard morphology assessment alone and with array cgh for good prognosis ivf patients: results from a randomized pilot study
Fertility Sterility
Generative adversarial nets
Automatic segmentation of the lumen region in intravascular images of the coronary artery
Med. Image Anal.
Segmentation of medical images using mean value guided contour
Med. Image Anal.
Cited by (46)
Exploring advanced architectural variations of nnUNet
2023, NeurocomputingA comparative study of attention mechanism based deep learning methods for bladder tumor segmentation
2023, International Journal of Medical InformaticsLWMA-Net: Light-weighted morphology attention learning for human embryo grading
2022, Computers in Biology and MedicineCitation Excerpt :However, previous methods involved complex mathematic algorithms and required manual feature extraction and selection process, which were time-consuming and limited in clinical applications. DCNNs have made impressive success in computer-assisted clinical applications, which include classification [10–12], segmentation [13,14], registration [15] and reconstruction [16] of medical images. However, utilization of DCNNs for human embryos assessment remains challenging, which may be due to the following reasons: First, tagging of individual embryos requires highly experienced embryologists, and is very time-consuming.
RU-Net: An improved U-Net placenta segmentation network based on ResNet
2022, Computer Methods and Programs in BiomedicineA coarse-to-fine full attention guided capsule network for medical image segmentation
2022, Biomedical Signal Processing and ControlCitation Excerpt :To alleviate the information loss in the U-Net, Yang et al. [28] replaced the convolution operations with dilated multiresidual blocks. In contrast, inceptioned [29] and ensemble [30] architectures were also developed for promoting the representation quality of the U-Net. Aiming at solving the optimal depth search and the identical-scale feature aggregation issues on the impact of the feature encoding performance of the U-Net, Zhou et al. [31] proposed an improved UNet++ formulation, which comprised an ensemble of U-Nets with varied depths and augmented with cross-scale feature aggregation mechanisms.