Approach to glaucoma diagnosis and prediction based on multiparameter neural network

To investigate the effect of comprehensive factor analysis on the relationship between glaucoma assessment and combined parameters including trans-laminar cribrosa pressure difference (TLCPD) and fractional pressure reserve (FPR).

The clinical data of 1029 patients with 15 indicators from the medical records of Beijing Tongren Hospital and 600 cases with 1322 indicators from Beijing Eye Research were collected. The doc2vec method was used to vectorize. The multivariate imputation by chained equations (MICE) method was used to interpolate. The original data combined with TLCPD, combined with FPR, and not combined parameters were respectively applied to train the neural network based on VGG16 and autoencoder to predict glaucoma and to evaluate the effect of combined parameters.

The accuracy rates used to classify the glaucoma of the two sets reach over 0.90, and the precision rates reach 0.70 and 0.80 respectively. After using TLCPD and FPR for the autoencoder method, the accuracy rates are both close to 1.0, and the precision rates are 0.90 and 0.70 respectively.

Using the combined parameters of FPR and TLCPD can effectively improve the diagnosis and prediction of glaucoma. Compared with TLCPD, FPR is more suitable for improving the effect of neural network for glaucoma classification.