Abstract
Supplementary Table S7. Illustration of each deep-learning model’s best prediction performance on SRH images with ≥130 high-quality patches (first model) and >140 high-quality patches (second and third model) after stepwise metric performance evaluation. The second model was used to predict the class of each acquired entity from small biopsy tissue samples. The level of accuracy depended on the final integrated diagnosis and the predefined model label assignment to the respective subclasses.