We developed a deep stacked undercomplete autoencoder (i.e., supervised) network to denoise the noisy 3D sectional images. Results demonstrate the feasibility of our proposed model in terms of peak-signal-to-noise ratio.