During thermal treatment of starch-based liquid food products, many coupled phenomena occur at different scales. Heat transfer affects local temperature evolution, which determines starch transformation (swelling and eventually amylose release and/or granules breakage). This transformation modifies the rheology, which in turn affects the fluid flow and heat transfer. This two-way coupling of flow, transfer and transformation was previously simulated in heat exchangers by Computational Fluid Dynamics (CFD), through purely Eulerian or Eulerian/Lagrangian approaches, assuming that the starch granules follow locally the fluid velocity. Moreover, the transformation was only characterized by a deterministic model for the mean particle size evolution (swelling). The present work takes into account directly the trajectory and the swelling of each starch granule in interaction with the surrounding fluid through an approach combining CFD with Discrete Element Method (DEM) simulations. Due to lubrication forces, the granules tend to avoid one another and when they come in contact, an elastic rebound is assumed. This approach is applied here to a waxy maize starch suspension during heating in a Couette rheometer. A shear-induced random walk of the particles is observed. Swelling first occurs near the heated wall, then swollen granules migrate toward zones of lower solid fraction. This multiscale model (rheometer scale: 1 mm, starch granule scale: 10 µm) can be used for the analysis of such rheometry tests in which the sample cannot be assumed as homogeneous.