Brugada syndrome (BrS) is a severe disease since it can cause nocturnal sudden cardiac death (SCD) without any other evident symptoms. BrS is usually diagnosed by finding a specific pattern in the electrocardiogram (ECG). Unfortunately, the only diagnostic pattern, the Type 1 pattern, is often hidden and can be revealed only after the administration of specific drugs as the ajmaline, which can be dangerous for the life of patients. To this purpose, recent research activities are focused toward the exploitation of deep learning approaches that are able to unmask the BrS without the use of such drugs. The aim of this paper is to introduce a deep learning approach for the identification of BrS based on the fusion of the phase information with the time domain signal. Although neglected in many signal processing applications, the exploitation of phase information has often produced a better outcome. Numerical results, evaluated on a real-world dataset, show the effectiveness of the proposed approach by obtaining, on a 7-fold cross-validation procedure, an averaged accuracy of about 85.2%.