Advantages of classical pancreatic image processing methods over machine learning (AI) techniques

Abstract

In the context of the increasing importance of automated analysis in the field of biomedical imaging, this study is aimed at a comparative assessment of traditional image processing algorithms and modern machine learning methods in the context of pancreatic pathology diagnostics. This organ is particularly difficult to segment due to its anatomical features and low tissue contrast. Despite the dominance of neural network technologies in modern scientific research, the work demonstrates that traditional methods such as threshold processing, active contours, and geometric analysis remain important in scenarios where processing speed, interpretability of results, or limited computing resources are critical factors. It is confirmed that artificial intelligence algorithms, although they demonstrate superior accuracy when working with heterogeneous tissues, require significant computational costs at the learning stage and depend on the quality of the initial data markup. At the same time, classical methods, without having a "black box" of neural network models, provide transparency, which allows medical specialists to adjust the processing parameters in real time. The article substantiates the expediency of using hybrid strategies combining preprocessing using classical methods and subsequent analysis using artificial intelligence technologies.