In the diagnosis of pediatric pneumonia, the weak conductivity contrast of lung tissues leads to limited image resolution in electrical impedance tomography (EIT). To improve the quality of image reconstruction, this study proposes a radial basis function neural network optimized by the crested porcupine optimizer and adaptive moment estimation (CPOA-RBFNN). By integrating the global search capability of swarm intelligence with the adaptive characteristics of gradient based optimization, the proposed method enhances both imaging accuracy and robustness. A total of 23 000 simulated datasets of pediatric pneumonia were constructed for training and testing. The proposed method was compared with Tikhonov regularization and the conventional radial basis function neural network (RBFNN), and evaluated using root mean square error (RMSE), image correlation coefficient (ICC), structural similarity index (SSIM), and peak signal-to-noise ratio (PSNR). Results show that under a noise level of 50 dB, the proposed method achieves the lowest RMSE (0.098), the highest ICC (0.922), and relatively high SSIM (0.909) and PSNR (9.295). Further clinical validation demonstrates that CPOA-RBFNN provides superior structural fidelity and lesion distinguishability in pneumonia reconstruction. In conclusion, the proposed method offers an effective solution for non-invasive, high-precision imaging and bedside-assisted diagnosis of pediatric pneumonia.