This article was originally published here
Front Med (Lausanne). 2022 Jan 18;8:676461. doi: 10.3389/fmed.2021.676461. eCollection 2021.
BACKGROUND: Post-transplant renal function is of critical importance to kidney transplant recipients. Accurate prediction of graft function would greatly help in deciding whether to accept or reject allocated kidneys.
METHODS : : Whole slide images (WSI) of H&E-stained donor kidney biopsies at ×200 magnification between January 2015 and December 2019 were collected. The clinical characteristics of each corresponding donor and recipient were retrieved. Graft function was indexed with stable estimated glomerular filtration rate (eGFR) and reduced graft function (RGF). We used convolutional neural network (CNN) based models, such as EfficientNet-B5, Inception-V3 and VGG19 for the prediction of these two outcomes.
RESULTS: A total of 219 recipients with H&E stained slides of donor kidneys were included for analysis [biopsies from standard criteria donor (SCD)/expanded criteria donor (ECD) was 191/28]. The results showed notable improvements in the prediction performance of the deep learning algorithm and the clinical feature model. The EfficientNet-B5 plus clinical data model showed the lowest mean absolute error (MAE) and root mean square error (RMSE). Compared to the clinical data model, the area under the receiver operating characteristic (ROC) curve (AUC) of the clinical data plus image model for eGFR classification increased from 0.69 to 0.83. In addition, the predictive performance of the RGF increased from 0.66 to 0.80. Gradient-weighted class activation maps (Grad-CAM) showed that the models localized areas of the tubules and interstitium near the glomeruli, which were discriminating features for RGF.
CONCLUSION: Our results preliminary show that deep learning for paraffin-embedded, formalin-fixed H&E-stained WSI improves the accuracy of graft function prediction for donor kidney transplant recipients deceased.