Comparative Analysis of Deep Learning Models for Classifying Ulcerative Colitis Stages on the Mayo Scale

Ulcerative colitis (referred to as UC from now on) is a chronic inflammatory bowel disease, which shows up throughout patients' lives in periods of activeness and periods of remission, constantly impacting quality of life. Classification regarding the severity of UC will help ensure proper and prompt treatment, improving patient outcomes. The three grades of the Mayo scoring system include Grade 1 (mild), Grade 2 (moderate), and Grade 3 (severe), which are widely used in clinical practice to classify UC. Cases can also be classified as intermediates between two grades if they do not clearly fit within a single score. However, there is inconsistency in the manual classification of endoscopic images, leading to variability between observers. Recent advances in machine learning, especially deep learning, have the potential to provide valuable tools to automate this procedure and enhance the accuracy of diagnosis. In this paper, we propose a deep learning-based method for the classification of UC severity from endoscopic images. We apply transfer learning techniques to address the standard limitation of small datasets in medical imaging, allowing our model to capitalize on pre-trained networks and focus its resources on distinguishing the visual features that indicate UC severity. Our approach demonstrates exceptional performance in classifying UC into the three Mayo grades and their intermediates, which is essential for timely intervention and better recovery rates.