Image-to-Image Retrieval
Users upload an image region of interest to retrieve similar image patches from the H&E collection from the NCI Lab of Pathology, TCGA, or CPTAC cohorts. HERE encodes the input image into a vector with the encoder and attention map. Then, the system will return top-K (50 by default) most similar image patches and their locations in the corresponding whole-tumor slides, together with clinical information if available.
1. You can scroll the whole-slide image on the left panel to examine the image details.
2. You can click the patches on the right panel to see the whole slide H&E image.
In the result page, the distance means the mininum and maxinum Euclidean distances between the query patch and retrieved patches. The rank_sum means the sum of rank scores computed from the background patches. Please refer to our manuscript for details.
2. You can click the patches on the right panel to see the whole slide H&E image.
In the result page, the distance means the mininum and maxinum Euclidean distances between the query patch and retrieved patches. The rank_sum means the sum of rank scores computed from the background patches. Please refer to our manuscript for details.
| image Show case-level prediction Show Case Info | attention map | retrieved similar images Help |