Automated Recognition of Zygote Cytoplasmic Area (ZCA) in Time-Lapse Imaging (TLI) Based on Deep Convolutional Neural Network (CNN)
Refereed conference paper presented and published in conference proceedings


摘要OBJECTIVE: Zygote morphology assessment may be an important tool to predict embryo development. With the technology of TLI, a series of images of zygotes can be captured; however, assessing their morphology parameters manually is still tedious. Several computer-assisted methods have been proposed recently, yet their effectiveness remains limited on oocyte or zygote morphology assessment. Thus, it is important to recognize ZCA accurately, and improve automated-assessment of morphology parameters. We sought to develop a computerized system based on deep CNN for the automated recognition of ZCA from images captured by TLI.
DESIGN: non-comparative; descriptive MATERIALS AND METHODS: We collected 520 images from 10 precleavage zygotes from a mouse embryo TLI database (1) as our dataset. The ZAC in each image was labeled by two experienced embryologists. We adopted and modified the deep CNN (2) to construct our ZCA recognition system. Two experiments (Exp.) were conducted to evaluate the performance of the processed system. 1. 5-fold cross-validation was performed in order to obtain the overall performance of our system. 2. Images were divided into two groups (obstacle and non-obstacle) based on whether there were blocks on the edge of the ZCA and to test our system’s performance on the obstacle group. All the recognition results are evaluated by the Intersection over Union (IoU). The IoU (a, b) between the predicted area a and the ground truth b is defined as IoU (a,b) ¼ jaXbj / jaWbj.
RESULTS: Detailed results can be found in Table I. Our results indicated that the ZAC of all images could be recognized with a high accuracy rate (>90%). In Exp. 1, the overall performance was 96.08%, and Exp. 2 showed our system could eliminate the effect of foreign bodies (96.27%).
CONCLUSIONS: We proposed a novel computerized system for automated ZCA recognition. We suggest that this system can be a robust tool in improving automated-assessment of zygote morphology parameters.
著者Zhao MP, Li H, Shi X, Chan YL, Luo X, Li TC
會議名稱73rd Scientific Congress and Expo of the American-Society-for-Reproductive-Medicine (ASRM)
會議地點San Antonio, Texas
會議論文集題名Fertility and Sterility
頁次E239 - E239

上次更新時間 2022-16-01 於 00:37