Deep Learning for Helicobacter Pylori Classification: A Comparative Approach Using Transfer Learning and Data Augmentation

• François F. R. Barbosa UFPI / IFMA
• Rodrigo N. Borges UFPI
• Patrick Ryan S. Santos UFMA
• Antonio Oseas C. Filho UFPI / IFMA
• Ivan S. Silva UFPI / IFMA
• Rodrigo M. S. Veras UFPI / IFMA

Resumo

Gastric cancer ranks among the most common types of carcinoma worldwide, with gastritis and Helicobacter pylori (HP) infections being its main precursors. Consequently, early diagnosis plays a crucial role in combating this disease. The use of Deep Learning (DL), particularly Convolutional Neural Networks (CNNs), has become widespread in research aimed at supporting disease diagnosis across various medical fields. In this study, we apply transfer learning and data augmentation techniques to compare five CNN-based models for the classification of HP, using a dataset comprising 204 histopathological images. Among the evaluated models, ResNet-101 achieved the best performance, with an accuracy of 94.61%, F1-score of 94.38%, precision of 97.00%, and recall of 92.29%. Our results indicate that, despite the limitations of the dataset, the proposed approach was effective, outperforming previous studies and reaffirming the potential of the applied techniques.