her2-reader / data /paper /paper-data /table_her2tasks.tex
nsugar's picture
HER2 reader: ViT-S/16 encoders at every stage, the deployed reader, per-slide predictions and paper data (part 3)
4367928 verified
Raw History Blame Contribute Delete
893 Bytes
% Generated by docs/paper/paper_her2tasks.py (TCGA-BRCA; 10 paired seeds)
\begin{tabularx}{\linewidth}{@{}>{\raggedright\arraybackslash}X r c cc@{}}
\toprule
& & HER2-low vs HER2-zero & \multicolumn{2}{c}{FISH amplification, read by the HER2 reader} \\
\cmidrule(lr){3-3}\cmidrule(lr){4-5}
tile encoder & params & dedicated head & Yale-trained & TCGA-trained, out of fold \\
\midrule
Phikon & 86M & 0.741$^{1}$ & 0.619$^{6}$ & 0.654$^{3}$ \\
H-optimus-0 & 1.1B & 0.692$^{2}$ & 0.706$^{1}$ & 0.690$^{1}$ \\
Hibou-B & 86M & 0.678$^{3}$ & 0.672$^{3}$ & 0.572$^{5}$ \\
\textbf{Ours (final)} & 22M & \textbf{0.649$^{4}$} & \textbf{0.682$^{2}$} & \textbf{0.669$^{2}$} \\
Midnight-12k & 1.1B & 0.626$^{5}$ & 0.650$^{4}$ & 0.644$^{4}$ \\
ResNet-50$^\ast$ & 26M & 0.615$^{6}$ & 0.575$^{7}$ & 0.445$^{7}$ \\
DINOv2-S$^\ast$ & 22M & 0.613$^{7}$ & 0.632$^{5}$ & 0.469$^{6}$ \\
\bottomrule
\end{tabularx}