Yashvardhan Jain1*, Leah L. Godwin1, Yingnan Ju1, Naveksha Sood1, Ellen M. Quardokus1, Andreas Bueckle1, Teri Longacre2, Aaron Horning 3, Yiing Lin4, Edward D. Esplin5, John W. Hickey6, Michael P. Snyder 5, N. Heath Patterson7, Jeffrey M. Spraggins8, Katy Börner 1*
All HuBMAP tissue datasets used in this study can be explored in their three-dimensional size, position, and rotation in the context of the Human Reference Atlas. 10 Datasets come from the left male, five from the right male, seven from the left female and eight from the right female kidney; four come from the male colon and three from the female colon. Using the Exploration User Interface (EUI), you can:
The colon WSI that are used for the prediction task are shown in a tissue viewer interface. The tissue section HBM462.JKCN.863 (see below) has 124 and HBM438.JXJW.249 (see below) has 36 crypts in the ground truth (GT) given in yellow. See legend in top right for color coding used for segmentation masks predicted by the five winning algorithms. For number of FTUs in solutions predicted by the five algorithms, see Supplementary Table 5 (Algorithm Performance).
Turn on and off different segmentations masks via selection of algorithms in legend. Hover over segmentations in tissue viewer to see WSI data without occlusion.
This interactive violin plot shows performance values for all five algorithms for three metrics: Dice (left), precision (top right), and recall (bottom right). For each metric, we show distribution for the ten kidney WSI with 2,038 glomeruli on the left and the distribution for the two colon WSI with 160 crypts transfer learning predictions on the right.
Run time performance was recorded for the training phase on kidney data, colon data exclusively (no transfer), and on kidney data and colon data, see Table 1. We also report run time for the two prediction tasks: from scratch without transfer learning (i.e., trained on five colon, tested on two colon datasets) and transfer learning (i.e., trained on 15 kidney datasets initially and then trained on five colon datasets, then tested on two colon datasets), see Methods section for details.