Automatic evaluation of human oocyte developmental potential from microscopy images

Abstract

Infertility is becoming an issue for an increasing number of couples. The most common solution, in vitro fertilization, requires embryologists to carefully examine light microscopy images of human oocytes to determine their developmental potential. We propose an automatic system to improve the speed, repeatability, and accuracy of this process. We first localize individual oocytes and identify their principal components using CNN (U-Net) segmentation. We calculate several descriptors based on geometry and texture. The final step is an SVM classifier. Both the segmentation and classification training are based on expert annotations. The presented approach leads to the classification accuracy of 70%.

Segmentation of mature human oocytes provides interpretable and improved blastocyst outcome predictions by a machine learning model

“…Another group described a pipeline where images of oocytes were first segmented into five classes (background, ooplasm, ZP, polar body, and residual cumulus cells), from which 24 features were extracted, and finally, feature vectors were used to classify oocytes as viable or nonviable (defined as the ability of the oocyte to become a well-developed embryo) 25 . This was the only study involving automatic segmentation of the oocyte that assessed the link between extracted features and clinical outcomes, with the most significant feature being the number of polar bodies present.…”

Fast Learning from Label Proportions with Small Bags

“…We focus on the particular case of small bags containing up to about 10 instances. As we will see, this scenario leads to better results by considering all possible configurations consistent with the annotations explicitly, and thus avoiding the otherwise necessary approximations, while still having some interesting applications, such as the embryo classification for IVF [2,3], or analyzing group photos in Section 3.…”

“…The first applications of LLP were in modeling voting behavior from aggregated electoral district data, where individual data is not available because of privacy requirements. In in vitro fertilization (IVF), the task is to predict the likelihood of successful development of individual embryos [2] or oocytes [3], while the only hard data is the outcome of the pregnancy, which may have resulted from any of the several implanted embryos. In Langerhans islet detection [4], only subjective classification of individual objects is possible, while the total contents in the sample can be quantified by DNA content measurement.…”