Myocardial infarction (MI) initiates pathological remodeling and alters myocardial stiffness. Accurate estimation of scar stiffness in MI is essential for assessing disease progression and risk stratifying MI patients. Traditional methods, such as pressure-volume loop analysis, are invasive and do not capture regional stiffness variations. This study presents a methodology for integrating cardiac strain data with computational modeling and deep learning (DL) to estimate myocardial scar stiffness non-invasively. The computational model simulated strains across healthy and infarcted myocardium to train a transfer learning-based DL model. The DL model demonstrated high predictive accuracy for mild MI cases (92.79%) and moderate accuracy for severe MI cases (88.68%) on test data. Higher accuracies were obtained for basal and mid-regional scar regions, whereas the accuracies for scars in apical and apex regions tended to be lower. The inclusion of scar location and severity highlighted the importance of comprehensive training on geometrical and biomechanical features to optimize accuracy. Indeed, data augmentation and transfer learning were employed to enhance the model’s generalizability. Our proposed framework offers a potentially non-invasive and clinically feasible approach to myocardial scar stiffness estimation, supporting the improvement of longitudinal monitoring and prognosis in MI patients. Future work will expand datasets and incorporate additional strain features to improve accuracy, particularly for complex remodeling scenarios.