Splitting the cortical surface into regions with homogeneous dMRI-based connectivity profiles is a promising but challenging topic. This paper extends the inter-subject connectivity-based cortex parcellation framework proposed by Roca [1]. In a first step, we implement the state-of-the-art algorithm with tuned parameters and, then propose a refined algorithm validated on the large high quality ARCHI database. This algorithm consists in clustering and subdividing each gyrus, in a reasonable time. Cross-validation shows that the resulting patterns are reproducible across groups. The stability is illustrated for the post-central gyrus of three different groups of subjects. Finally, the method has successfully been made robust to initial conditions: tracking type, cortical mesh characteristics and boundaries of gyri.