Learning Interpretable representation in medical applications is becoming\nessential for adopting data-driven models into clinical practice. It has been\nrecently shown that learning a disentangled feature representation is important\nfor a more compact and explainable representation of the data. In this paper,\nwe introduce a novel adversarial variational autoencoder with a total\ncorrelation constraint to enforce independence on the latent representation\nwhile preserving the reconstruction fidelity. Our proposed method is validated\non a publicly available dataset showing that the learned disentangled\nrepresentation is not only interpretable, but also superior to the\nstate-of-the-art methods. We report a relative improvement of 81.50% in terms\nof disentanglement, 11.60% in clustering, and 2% in supervised classification\nwith a few amounts of labeled data.\n