Radiofrequency (RF)-induced heating presents significant safety concerns for patients with medical implants undergoing magnetic resonance imaging (MRI). This study proposes a six-layer feedforward artificial neural network (ANN) for rapid prediction of RF-induced heating near thoracic implants. Anatomical variability was incorporated through three high-resolution human models (Ella, Duke, FATS) with six implant sampling regions (L1–L3, R1–R3). A total of 2,592 electromagnetic simulations were performed under 1.5 T MRI conditions, from which six input features were extracted to predict two critical heating metrics: AvgEtotBelowLimit and AveEtot. All training data were normalized to a whole-body SAR of 2 W/kg. The ANN converged within 100 epochs and demonstrated exceptional test performance (n = 260), with MSE/R2 values of 150.51/0.9912 for AvgEtotBelowLimit and 179.22/0.9890 for AveEtot. Error distributions centered within [−20, +20] confirmed robustness across anatomical variations. Scatter plots showed strong linear alignment (y = x) with outliers (<2%) primarily at tissue interfaces. The ANN reduces prediction time from hours to milliseconds while complying with IEC/ISO safety margins, enabling real-time MRI safety assessment for patients with thoracic implants.