Scroll wave chaos is thought to underlie life-threatening ventricular fibrillation. However, currently there is no direct way to measure action potential wave patterns transmurally throughout the thick ventricular heart muscle. Consequently, direct observation of three-dimensional electrical scroll wave chaos remains elusive. Here, we study whether it is possible to reconstruct simulated three-dimensional scroll wave chaos inside a bulk-shaped excitable medium from two-dimensional observations of the wave dynamics on the bulk's surface using deep learning. We trained encoding-decoding convolutional neural networks to predict three-dimensional scroll wave chaos inside opaque and transparent as well as isotropic and anisotropic excitable media from two-dimensional projections or observations of the wave dynamics on the surface. We tested whether observations from one or two opposing surfaces would be sufficient, whether incorporating measurements of the surface deformation improves the reconstruction, and tested the feasibility of predicting the bulk's thickness. We demonstrate that it is possible to fully reconstruct three-dimensional scroll wave chaos in transparent excitable media with anisotropy and to obtain partial reconstructions in opaque excitable media when analyzing two opposing layers of the bulk. We found that anisotropy provides crucial information for neural networks to decode depth, which facilitates the reconstructions. In the future, deep neural networks could be used to visualize transmural action potential wave patterns during ventricular fibrillation from epi- or endocardial recordings.