<p>Accurately predicting the fatigue life of additive manufactured metals remains challenging due to complex defect morphology and failure mechanisms. This work proposes a critical directional curvature feature of volumetric defects for nondestructive fatigue life prediction in laser powder bed fusion (L-PBF) Ti-6Al-4V parts via data-driven models. This feature is defined as the maximum curvature component of the defect surface along the loading direction and is extracted from X-ray computed tomography to capture the local geometric sharpness with respect to the loading direction. To build the physical foundation of this feature, destructive fractography analysis is performed using a defined critical contour curvature feature, revealing a strong negative correlation between curvature-based features and fatigue life. The findings show that sharper local geometries with higher curvature increase local stresses, accelerate crack initiation, and reduce fatigue life. The critical contour curvature feature demonstrates a strong correlation (Pearson = 0.854, Spearman = 0.836) with the critical directional curvature, indicating consistent defect characterization in both destructive and nondestructive analysis. Furthermore, the results show that incorporating this feature substantially improved fatigue life prediction accuracy by 22.46% in linear regression and 35.34% in neural network after removing stress amplitude effect, compared to the baseline with only conventional defect size and location features. Additionally, the Shapley Additive Explanations and finite element simulation are employed to explain the neural network’s predicted outcomes and validate the relevance of the curvature-based feature in relation to local stress concentration. This approach advances nondestructive fatigue life prediction and defect criticality analysis in L-PBF metals through data-driven modeling and physics-based interpretation.</p>