Objective  To explore a method for rapid and differential diagnosis of rejection after kidney transplantation through urine hyperspectral imaging technology.

Methods  Hyperspectral data information from urine samples of 118 recipients after kidney transplantation was collected, and a deep learning model was constructed to diagnose and classify the types of rejection.

Results  A deep learning diagnostic model based on the 34-layer residual network (ResNet-34) was constructed, and 118 patients were included and divided into the training set and the test set. Based on the pathological results of the transplanted kidney puncture, the urine samples of the patients were classified into five groups: the non-rejection group, the T-cell-mediated rejection group, the antibody-mediated rejection group, the mixed rejection group and the nephropathy recurrence group. The results showed that the diagnostic sensitivities of the model for the above five groups were 0.960, 0.980, 0.930, 0.940 and 0.943 respectively, and the diagnostic specificities were 0.983, 0.993, 0.997, 0.989 and 0.989 respectively. The overall diagnostic accuracy rate reached 95.7%.

Conclusions The study provides a non-invasive, rapid and accurate auxiliary diagnostic method for the differential diagnosis of rejection after kidney transplantation.