Table 7 Comparison between references (expert pathologists), and our method on slide-level tasks using the EfficientNet-B3 architecture, including the consensus ground truth (with 8% of the dataset removed).
From: Overcoming the limitations of patch-based learning to detect cancer in whole slide images
Reference | Target | Dice (% over-lap) | Error in \(\sqrt{{d}_{1}{d}_{2}}\) | Cfs. mtx |
|---|---|---|---|---|
P#1 | CNN | 64 | 2.12 | \(\left(\begin{array}{cc}6& 0\\ 0& 20\end{array}\right)\) |
P#2 | 74 | 2.20 | \(\left(\begin{array}{cc}6& 0\\ 0& 20\end{array}\right)\) | |
P#3 | 79 | 1.75 | \(\left(\begin{array}{cc}6& 0\\ 0& 20\end{array}\right)\) | |
P#2 | CNN | 66 | 3.06 | \(\left(\begin{array}{cc}6& 0\\ 0& 20\end{array}\right)\) |
P#3 | 81 | 1.63 | \(\left(\begin{array}{cc}6& 0\\ 0& 20\end{array}\right)\) | |
P#3 | CNN | 67 | 1.86 | \(\left(\begin{array}{cc}6& 0\\ 0& 20\end{array}\right)\) |
Consen-sus | CNN | 66 | 2.33 | \(\left(\begin{array}{cc}6& 1\\ 0& 19\end{array}\right)\) |
P#1 | 73 | 1.65 | \(\left(\begin{array}{cc}6& 1\\ 0& 19\end{array}\right)\) | |
P#2 | 88 | 0.79 | \(\left(\begin{array}{cc}6& 1\\ 0& 19\end{array}\right)\) | |
P#3 | 79 | 1.48 | \(\left(\begin{array}{cc}6& 1\\ 0& 19\end{array}\right)\) |
