This paper presents a comparative study of robust diffusion algorithms when used for smoothing structural fields applied in volumetric image interpolation. The input data consists of a set of parallel and equidistant slices which are considered sparsely located along a central axis. The structural flows are constructed using the dual directional block matching algorithm (DBMA). Two vectorial flows are modelled in both directions along the central axis, using the correlation between blocks of pixels from successive slices. As with most block matching algorithms, this method is susceptible to noise and the resulting vector fields contain outliers. A methodology that combines diffusion and robust statistics in order to smooth the dual structural flow is proposed. Consequently, new slices are interpolated in between the existing slices, according to the smoothed vector fields. The set of algorithms is applied in volumetric medical images.