Fig. 2: Model architecture and predictive performance.

The figure illustrates the model framework and compares the predictive performance between the proposed model and baseline models. a Architecture of the early screening model. b ROC curves of the proposed model (FP + MPI), its variant (FP + CRF), and baseline models including MPI-based model, FRS, PCE, QRISK3, SCORE, RMF, RMF + MPI, and FP-based model. c Calibration curves. The x-axis represents the mean predicted values and the y-axis represents the fraction of positive samples. Due to the significant imbalance between positive and negative samples in the internal validation set, the calibration curve was plotted by under-sampling the negative samples to achieve a 1:1 ratio with the positive samples, using the diagonal line as a reference. d Decision curves. The x-axis represents the threshold probability and the y-axis represents the net benefit. e Kaplan–Meier graphs for major vascular-related events in low-risk, middle-risk, and high-risk groups. AUC area under the receiver operating characteristic curve, BS Brier score, CRF clinical risk factor, FP fundus photograph, FRS Framingham risk score, MPI minimal personal information, PCE pooled cohort equation, RMF retinal microvascular features, SCORE systematic coronary risk evaluation.