Fig. 4: Results of the four experiments on the MGH dataset.

(a) Validation of CVD screening. b–d Comparison of our deep learning model with three clinically used risk scores, including (b) CAC score³⁴, (c) CAD-RADS³⁵, and (d) MESA 10-year risk score³⁶. p values were computed using a one-tail z-test.