The Santos Basin is the most prolific pre-salt oil-producing offshore basin in Brazil. The basin has experienced several magmatic events and volcanic rocks have been found in Pre-salt exploration and appraisal wells. The presence, thickness and spatial extent of these volcanic rocks are challenging to predict from seismic, stemming from (i) the current degree of understanding the geometry and genesis of these volcanics, (ii) seismic quality issues and (iii) the limited well control. Typically, volcanics can act as barriers and can cause diagenetic effects in the carbonate reservoirs. In our study, two pre-salt carbonate reservoirs (Itapema and Barra Velha Formations) were investigated. These heterogeneous carbonates are located beneath a very thick salt layer (Ariri Fm.). The main reservoir consists of several hundreds of meters of bioclastic calcirudite facies (coquinas) and organic rich shales of the Itapema Fm., while the Barra Velha Fm. consists of in-situ facies (shrubstones and spherulestones) and reworked facies (calcarenite and calcirudite). Volcanic intrusions more than 500 meters in thickness have been found in both reservoirs. Petrophysical analysis established that these intrusive volcanic intervals typically display high bulk density, high velocity (hence high acoustic impedance), low gamma-ray and low neutron porosity values. The 3D seismic PSDM volume underwent a fit-for-purpose seismic data conditioning, in particular seismic noise cancelation and frequency enhancement, steered by zero-offset vertical-seismic profile data. Detailed mapping of faults using geometric seismic attributes and spectral decomposition, gravity and magnetic data revealed a complex tectonic setting including normal, strike-slip and regional transform faults. Manual fault mapping, 3D seismic interpretation and seismic facies analysis suggest that magmatism appears to be structurally controlled, possibly by deep seated faults. A deep neural network based acoustic impedance inversion, calibrated using several wells, conventional- and sidewall core samples. We adopted a neural network (NN) approach in order to (i) capture possible non-linear relationship between input data and targeted reservoir property and (ii) to generate higher vertical resolution volumes, which is attractive to image thin layers. This inverted volume was used to assist on identifying volcanic bodies from the 3D seismic, where volcanic cones, dikes, sills and probably laccoliths can be interpreted.