Optimal Cut-Off Points for Pancreatic Cancer Detection Using Deep Learning Techniques

Deep learning-based approaches are attracting increasing attention in medicine. Applying deep learning models to specific tasks in the medical field is very useful for early disease detection. In this study, the problem of detecting pancreatic cancer by classifying CT images was solved using the provided deep learning-based framework. The choice of the optimal cut-off point is particularly important for an effective assessment of the results of the classification. In order to investigate the capabilities of the deep learning-based framework and to maximise pancreatic cancer diagnostic performance through the selection of optimal cut-off points, experimental studies were carried out using open-access data. Four classification accuracy metrics (Youden index, closest-to-(0,1) criterion, balanced accuracy, g-mean) were used to find the optimal cut-off point in order to balance sensitivity and specificity. This study compares different approaches for finding the optimal cut-off points and selects those that are most clinically relevant.