Fig. 4
Significant performance differences between OpenSlide extracted original MRXS and DICOM converted images exist during AI model inference. (a) AI models (ResNet18, cTransPath, Virchow, and Virchow2) were trained on TCGA data and tested on local SIP data for the OpenSlide extracted original MRXS and DICOM converted images for different resolutions and data set sizes. Exemplary results are shown in (b–d). Results in (b) belong to the cTransPath model trained with a small subset (n = 17) from site A (ovarian cancer) at a resolution of 1 mpp. Results in (c,d) belong to the Virchow model trained with a small subset from Site A and B (n = 17) from bladder data at a resolution of 1 mpp.
