Table 1 A comparison between different methods that one might be interested in regarding accuracy and efficiency. Directly classifying downsized WSIs is very fast, but the accuracy is unacceptable. Exhaustively predicting every patch at every zoom level brings a minimal increase in accuracy but doubles the average time. Our selective multi-zoom-level patch prediction method ensures efficiency without compromising accuracy.
From: Deep learning system for true- and pseudo-invasion in colorectal polyps
Method | Accuracy (%) | Average time (s) |
|---|---|---|
Directly classifying downsized WSIs | 59.7 | 0.3 |
Exhaustive multi-zoom-level prediction | 84.2 | 1487.1 |
Selective multi-zoom-level prediction (ours) | 83.9 | 771.5 |
