Evidence fusion with contextual discounting for multi-modality medical image segmentation