Fig. 3: CTC-Tracer can map CTCs to the primary tumor atlas efficiently.
From: Deep transfer learning enables lesion tracing of circulating tumor cells
a 2D visualization of primary tumor cells (50318 cells) involved in this study. The full name of these cancers can be found in Supplementary Table 5. b Changes of loss function values (detailed in Eqs. 7,16–18 in Methods, including the loss function l, and the three loss items included in the loss function lreg, lcdd and lce) throughout the entire training process. c Changes of prediction accuracy during the CTC-Tracer training process. d Changes of prediction accuracy on various CTC datasets (including MEL, HCC, BRCA and PC CTCs) during the CTC-Tracer training process. e 2D embedding of primary tumor samples and CTC samples (372 cells, 4 cancer types) before domain adaptation using t-SNE. CTCs and primary cancer cells from the same organ are discretely distributed. f t-SNE 2D embedding of primary tumor samples and CTC samples after domain adaptation. CTCs and primary cancer cells from the same organ are located together after domain adaptation.
