Introduction: X-ray based imaging is widely used for guiding endovascular aortic repairs (EVAR). Still, proper interpretation of hidden and obscured anatomy remains a challenge and often requires repeated imaging from several perspectives. Objective: Our aim is to decompose a conventional X-ray image into several independent, non-overlapping depth layers, such that the sum resembles the original input. Material and Methods: Conventionally, the creation of one single X-ray image X is guided by Beer-Lambert’s law, X=I0⋅exp(−∫0Lμ(x)dx), where I0 X-ray photons interact with human tissue of thickness L guided by the tissue attenuation function μ(x). The integral over thickness L can now be subdivided into k+1 ranges [0,d1],]d1,d2],]d2,d3],…,]dk,L] introducing thickness layers d1,…,dk (k∈ ¥). Setting k=2, we can change Beer-Lambert’s law to X=I0⋅exp(−∫0Lμ(x)dx)=I0⋅exp(−∫0d1μ(x)dx−∫d1d2μ(x)dx−∫d2Lμ(x)dx). Each subintegral now describes an X-ray image on its own (depth layer image):X1=I0⋅exp(∫0d1μ(x)dx),X2=I0⋅exp(∫d1d2μ(x)dx),X3=I0⋅exp(∫d2Lμ(x)dx)and hence, every X-ray image X can be reconstructed viaXt=f(X;{d1,…,dk})s.t.∑t=13Xt=Xwith an objective function f and corresponding thickness layers. As we integrate this objective function in a deep autoencoder scheme, we define f to minimize the reconstruction error ‖∑d=13Xd−X‖2while keeping individual depth layer images smooth and sparse. We have validated our proposed algorithm on clinical datasets of three patients (about 7200 X-ray images). Collections were divided into non-overlapping sets of training (60%), validation (20%), and testing (20%). Different metrics were computed to evaluate the performance of our model; Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) to measure the quality of reconstruction and perception, respectively. We further employed different statistic metrics for comparison to ground truth depth values; such as Mean Squared Error (MSE), Normalized Cross Correlation (NCC) and Relative Error (RE). Results: Our model yields a PSNR of 53.02% and a SSIM of 86.14. Findings on MSE, NCC and RE were 4.40% ± 2.04, 66.50% ± 13.94, and 23.93% ± 6.83 respectively. Our partner physicians have deemed visual results very promising. Summary and Conclusion: We have presented a new concept for enhancing depth perception for EVAR allowing revealing hidden and obscured structures without the need for repeated imaging. Despite the challenging aspects of modeling such a highly ill posed problem, exciting and encouraging results are obtained paving the path for further contributions in this direction.