Fig 1.
The oocytes were split into control and experimental groups to verify the safety of the oocyte after the aspiration measurement. Both groups followed the standard ICSI procedure to check the fertilization, good day 3 embryo, and blastocyst formation rate (see method of data analysis). Subsequently, the experimental group were further used to build a model to predict the usable blastocyst using biomechanical properties of oocytes (see method of determining the biomechanical properties of oocytes and data analysis). The accuracy of predicting usable blastocyst formation was compared between the model and the embryologists.
Table 1.
A split cohort study was conducted at two IVF clinics.
Fig 2.
Workflow for determining the biomechanical properties of the oocytes.
The workflow is divided into aspiration depth measurement, video analysis, and parameter extraction.
Fig 3.
Aspiration depth measurement procedure.
A small suction is applied to the oocyte through a micropipette. The procedure is recorded with a video camera at 70 frames per second.
Fig 4.
Aspiration depth video analysis and model fitting.
(a) The photo shows an oocyte that had been partially pulled into the micropipette. The aspiration depth was measured between the position of the zona pellucida and the tip of the micropipette. (b) The picture shows an example of the measured aspiration depth and the modeled aspiration depth of an oocyte.
Fig 5.
In the Video processing, ROI (region of interest) Cropping zooms in on the video to measure the inner boundary of zona pellucida, inner diameter of micropipette, and aspiration depth inside the micropipette. Start Frame Selection starts videos from a frame before the first moved frame. Zona Thickness Calculation calculates the thickness of zona pellucida as an initial aspiration depth. Inner Boundary of Zona Calculation determines the position of the inner zona pellucida near the micropipette tip. Aspirated Zona Tracking Calculation detects the position of zona pellucida inside of micropipette each frame over time. Time Vector keeps moved frames in the video and shrinks the video length to reduce the analysis time. Aspiration Depth Calculation calculates the aspiration depth of oocyte each frame over time. Conversion Factor Normalization standardizes the differences in video resolution by normalizing the aspiration depth with the inner diameter size of micropipette. In the Pressure processing, Applied Force Calculation measures the pressure force applied on the oocyte when aspirated into micropipette. The Modified Zener Model finds the mechanical parameters of viscoelastic behavior of oocyte as features of biomechanical properties for the predictive classifier. Forward Feature selection determines the best combination of features between clinical information and mechanical properties for model training. Modeling uses greedy search to find the best parameters for building a SVM classifier during cross-validation procedure. Predictive outcome provides a result indicating whether the blastocyst is usable or unusable.
Fig 6.
Outcomes from the Shenzhen Army Hospital.
Fertilization rates and day 3 good embryo rates of the control and the experimental groups. The fertilization rate was not significantly different between the two groups (p = 0.7282), while the day 3 good embryo rates were not negatively affected by the aspiration measurement in the experimental group (p<0.01).
Fig 7.
Outcomes from the Taiwan IVF Group Center.
Fertilization rate, day 3 good embryo rates and blastocyst formation rates of the control and the experimental group. The fertilization and day 3 good embryo rates were not significantly different between the two groups (p = 0.69 experimental and p = 0.63 control), the blastocyst formation rates were not negatively affected by the aspiration measurement in the experimental group (p<0.01).
Table 2.
Prediction results of the usable blastocysts predictive classifier with biomechanical properties compared to maternal factors and morphological information by four embryologists.