By aggregating visual word and sentence features, our Zig-RiR can effectively explore both global and local information while preserving spatial continuity. Experiments on four medical image ...