Endoscopy through coherent fiber bundles plays a significant role in industrial and medical 2D imaging. By replacing the lens on the distal side with a diffuser, the 3D information of the measurement volume is encoded as 2D speckle patterns on the camera. Neural Networks can then be employed to reconstruct the 3D object. Therefore, minimally invasive single-shot 3D imaging through a flexible low-cost endoscope with a diameter of less than 1 mm is enabled. However, the number of fiber cores is limiting the transferable information and reduces the reconstruction quality. In this paper, separate reconstruction for the diffuser and the coherent fiber bundle for different core numbers is explored. This approach enables biomedical applications for in vivo diagnostics, e.g. with fluorescence imaging in 3D.