Visual Interpretability of Capsule Network for Medical Image analysis

Deep learning (DL) models are currently not widely deployed for critical tasks such as in health. This is attributable to the "black box", making it difficult gain trust of practitioners. paper proposes use visualizations enhance performance verification, improve monitoring, ensure understandability, and interpretability needed practitioners' confidence. These demonstrated through development a CapsNet model recognition gastrointestinal tract infection. The comprises several organs joined long tube from mouth anus. It susceptive diseases that medics diagnose, since easy have physical access sick regions. Consequently, manual analysis images unhealthy parts requires skills an expert, tedious, prone errors, costly. Experimental results show form post-hoc can demonstrate reliability applied diseases. outputs also be explained practitioners? confidence paving way its adoption areas society.