Figure 3

MPI AI model performance in the testing cohort (n = 388). (A) The risk score derived from the baseline MPI images can stratify patients into below-median-risk group (shown in blue color, also referred to as the low-risk group) and above median-risk group (shown in red color), which have significant difference in their time to cardiovascular incidents (P < 0.0001). (B) The above-median risk group were further stratified into two equal-sized groups (n = 97 each), and referred to as the high (shown in green color, and corresponds to the 4th quartile) and intermediate (shown in red color, and corresponds to the 3rd quartile) risk groups respectively. The corresponding survival curves of the two groups have significant difference (log-rank P = 0.0023). The intermediate risk and the low risk groups, also manifest statistically significant difference (log-rank P = 0.0472). When the low-risk group (corresponds to the 1st and 2nd quartiles together, accounts for 50% of all patients) and the high-risk group (accounts for 25% of all patients) are compared, the difference in survival curves is very significant (log-rank P < 0.0001). (C) The Kaplan–Meier plots of patients stratified as single vessel disease (SVD), double vessel disease (DVD) and thriple vessel disease (TVD). (D) The ROC of clinical model, vessel number, clinical and vessel combined model, as well as the MPI score for MACE events. (E) The ROC for Total CV events. (F) The ROC of patients stratified using the stent number. (G) The forest plot illustrates hazard ratios (HR) and their 95% confidence intervals (CI) for various clinical and ICA factors on the association between MPI derived scores and cardiovascular incidents. Among 388 patients, the score showed a strong association (HR: 8.328, 95% CI 3.604–19.245; p < 0.001). Male patients (n = 295) exhibited a higher association (HR: 13.411, 95% CI 4.290–41.926; p < 0.001) compared to the overall population, while females (n = 93) displayed a significant but comparatively lower association (HR: 5.535, 95% CI 1.355–22.603; p = 0.017). Evaluation of blocked vessels revealed stronger associations in patients with more blockages (HR: 18.377 for three vessels, p < 0.001), while the presence of stents showed a rising trend in model performance, peaking notably in cases with three or more stents (HR: 57.446; p = 0.019) despite a smaller sample size.