The magnetic resonance (MR) radiofrequency (RF) -induced heating is one of the primary risks for patients with implanted shoulder prosthesis since its metallic part interacts with the radiated RF field and deposits the RF power in the tissue. To rapidly and accurately estimate the RF-induced heating of shoulder prosthesis, we develop a 5-layer feedforward artificial neural network (ANN) in the study. The detailed parameterized description of the shoulder prosthesis is adapted to the input features of this ANN model, including the diameter and length of the humeral stem and the diameter of the humeral head. In total, 150 distinct shoulder prosthesis configurations with varying dimensions are constructed. The electromagnetic (EM) numerical simulations are conducted, and 1g-averaged SAR (SAR1g) values are calculated to evaluate the RF-induced heating, which is further applied to the output of the ANN model. The ANN model achieves convergence within 300 training epochs with elaborate optimization. All the training data is normalized to whole-body SAR of 2 W/kg. Mean absolute errors (MAE) are 1.92 W/kg, 2.44 W/kg, and 1.86 W/kg for the training, validation, and test sets, respectively. The coefficient of determination (R2) approached 0.99, indicating high predictive accuracy and strong generalization ability. Consequently, this research can provide an efficient and reliable option for evaluating RF-induced heating of shoulder prosthesis.