The paper addresses the issue of overall survival prediction in head and neck cancer as an effective mean of improving clinical diagnosis and treatment planning. A new solution is proposed using semi-invertible convolutional networks. Our model exploits the 3D features of computed tomography (CT) scans to enrich the dataset used in the learning phase and thereby improve the prediction accuracy. This is achieved by designing a first architecture featuring a combination of a CNN classifier with a fully convolutional network pre-processor. The latter has been replaced in the second solution by an invertible network to deal with the memory constraints noticed in the first architecture. Obtained results showed that both architectures have led to considerable improvements in terms of prediction accuracy (0.75) compared to state-of-the-art solutions.