2.5D Deep Learning Approach for Precise Image Segmentation of Gastro-Intestinal Tract

This paper explores a novel 2.5D training approach for Gastro-Intestinal(GI) Image Segmentation where 3D data is trained as 2D images effectively by stacking consecutive magnetic resonance imaging (MRI) scans as channels. This approach minimizes computational and memory demands, simplifies pipelines, and improves inference efficiency over traditional 3D training methodologies. Empirical results on the dataset from the Kaggle competition - University of Wisconsin Madison gastrointestinal tract image segmentation demonstrate improved scores, attributed to the enriched depth information. This innovative method offers promising opportunities to advance medical image analysis while optimizing resource utilization.