TABLE 2.
System performance of each step of CNN‐based cell classification
| Bethesda classification | YOLOv4 performance (first step) | ResNeSt performance (second step) | ||||||
|---|---|---|---|---|---|---|---|---|
| Accuracy | Precision | Recall | F‐measure | Accuracy | Precision | Recall | F‐measure | |
| NILM | 77.0% | 100.0% | 49.0% | 65.8% | 98.0% | 92.3% | 92.3% | 92.3% |
| ASC‐US | 76.6% | 27.9% | 86.4% | 42.2% | 87.2% | 60.0% | 41.4% | 49.0% |
| LSIL | 90.1% | 64.3% | 34.6% | 45.0% | 92.0% | 76.2% | 59.3% | 66.7% |
| ASC‐H | 89.6% | 75.0% | 31.0% | 43.9% | 92.1% | 66.7% | 90.3% | 76.7% |
| HSIL | 89.2% | 52.3% | 88.5% | 65.7% | 85.5% | 57.9% | 61.1% | 59.5% |
| SCC | 91.9% | 51.4% | 94.7% | 66.7% | 88.4% | 77.6% | 80.4% | 78.9% |
| Average | 85.7% | 54.2% | 67.0% | 54.9% | 90.5% | 71.8% | 70.8% | 70.5% |