Table 2 Comparison of static microscopy (AI-based), dynamic microscopy (AI-based) and mechanical method for oocyte evaluation.
From: Machine learning and microfluidic integration for oocyte quality prediction
Observation method | |||
|---|---|---|---|
Static image-based AI21 | Time-lapse microscopy (TLM)-based dynamic analysis36 | Our proposed microfluidic-AI integrated method | |
Morphological features at a single point in time | Morphological changes over time | Mechanical properties of oocytes | |
Evaluated parameter | |||
Evaluated parameter | Morphological features from 2D image (e.g., solidity, circularity) | time to polar body appearance, time to pronuclei appearance and time to pronuclear fading | Diameter, Flow Q, Deformation Index |
Intracellular features | Not considered explicitly | Partially inferred from dynamic morphology | Considered |
Temporal resolution | Single image per oocyte | 72h TL with ~ 400–500 frames per embryo | Snapshot during channel passage (~ 2 min) |
Computational complexity | Lower | Higher | Lower |
Accuracy in selection | AUC = 0.67 | Accuracy = 93%, Sens = 97%, Spec = 77% | Accuracy = 76.1% (K-Fold), 75.9% (LOO) |
Equipment cost (USD) | ~ 5,000 (standard microscope + camera) | ~ 100,000 + (EmbryoScope + TLM system) | ~ 5,000 (pump + microscope + camera) |
Data volume | 11,757 oocyte images | 704 embryo TL videos | 54 oocyte trials |
Real-time analysis | No (manual post-processing) | Yes (automated TL analysis) | Yes (real-time tracking in chip) |
Training data requirement | Large (image-based, annotated) | Extensive (multi-frame labeling) | Moderate (single reading per oocyte) |
Susceptibility to noise | Low (clear still images) | High (lighting/motion variability) | Medium (flow fluctuation, debris) |
Implementation ease | Simple microscopy setup | TLM system required, expert setup | Custom chip + camera, moderate effort |
Monitoring duration | ~ 5–10 s per image | Up to 72h per embryo | ~ 1–2 min per test |
