Hyperspectral imaging to predict the effect of cyclophosphamide in primary membranous nephropathy.

Background: Presently, there is a lack of accurate predictors of the efficacy of primary membranous nephropathy. The aim of this study is to explore the application value of hyperspectral imaging in predicting the efficacy of cyclophosphamide treatment in primary membranous nephropathy.

Method: A total of 30 patients with primary membranous nephropathy who were treated with glucocorticoid combined with cyclophosphamide were collected. Hematoxylin-eosin stained renal pathological images were acquired by hyperspectral imaging system at the spectral range of 400-1000 nm. The remission group data set contained 23 samples, while the non-remission group data set contained 28 samples. A one-dimensional convolutional neural network model was established to train and test the hyperspectral data, and the performance of the model was evaluated.

Result: From the spectral curve, the spectral difference between the remission group and the non-remission group was obvious between 525 and 700 nm. The spectral data in this band were analyzed by one-dimensional convolutional neural network, and the confusion matrix showed that the remission group and the non-remission group were successfully classified. The precision and recall were 0.89 and 0.81 for the non-response group and 0.83 and 0.90 for the response group, respectively, with an F1 score of 0.85 in both groups. The area under the AUC curve of the classification model reached 0.857.

Conclusion: In this study, a one-dimensional convolutional neural network model was used to analyze the hyperspectral images of renal pathology of PMN patients, and the patients in remission group and non-remission group were successfully classified after glucocorticoid combined with cyclophosphamide treatment.