Figure 3 | Scientific Reports

Figure 3

From: Efficient prediction of a spatial transcriptomics profile better characterizes breast cancer tissue sections without costly experimentation

Figure 3

Super-resolution of SPARC expression as a practical application of DeepSpaCE. Super-resolution of SPARC expression using DeepSpaCE highlights tumor invasion more clearly and semi-supervised learning for tissue section imputation using DeepSpaCE improves prediction accuracy. (a) Nine images show the super-resolved results for SPARC expression. Three images in the left column show section D2 after H&E staining. Three images in the middle column show the heatmaps of predicted SPARC expression by DeepSpaCE for the original spots (top), imputed spots (middle), and both original and imputed spots (bottom). Three images in the right column show overlays of predicted SPARC expression by DeepSpaCE and H&E staining for section D2. (b) Three enlarged images on the right area show tumor cell invasion (blue arrow) and the microenvironment (red arrow). Spot size is adjusted to smaller than the exact spot size of the Visium platform to show the background image. (c) Left image shows the H&E-stained section adjacent to section D2. Right enlarged image is the same region as (b). Enlarged image shows the invasion of tumor cells. We drew annotated outlines of invasive regions in the H&E images. The red arrows show the invasive region.

Back to article page