The difficulty in early diagnosis of pancreatic ductal adenocarcinoma is one of the main reasons for its high mortality rate. To facilitate this diagnosis, different versions of three types of artificial intelligence and data approximation algorithms have been developed: genetic algorithm, neural networks, and logistic regression. These take data on sex, age, CA19-9 in blood, and creatinine, TFF1, REG1B, and LYVE1 levels in urine from hundreds of patients and classify them as control cases, those with a benign tumor, or those with a malignant one.

The accuracy, specificity, and sensitivity of these algorithms have been studied, determining which ones offer the greatest predictive capacity. These algorithms were compared with each other and with results obtained in previous literature. The genetic algorithm is the most novel. Its predictive capacity has proven to be comparable to those previously studied, achieving the highest accuracy for classifying controls and malignant tumors. Furthermore, sensitivities and specificities greater than 80% were obtained for all three methods for this classification.

This confirms the potential of machine learning tools for diagnosing this type of tumor, although there are still limitations to their clinical implementation.

The work is available in the repository .