Histopathological Gastric Cancer Detection Using Transfer Learning

Gastric cancer is a leading cause of cancer-related deaths worldwide, underscoring the need for early detection to improve patient survival rates. Histopathological image analysis (HIA) is the gold standard for this purpose but it is time-consuming and laborious, leading to interest in computer-aided diagnosis to assist pathologists. Deep learning has shown promise in various HIA tasks, but the limited amount of training data in medical imaging has posed a significant obstacle. In this paper, we propose transfer learning-based CNNs for binary classification of gastric cancer patches, overcoming this limitation. We validate our approach on the publicly available GasHisSDB dataset, which includes three sub-databases with patch sizes of 80 x 80 pixels, 120 x 120 pixels, and 160 x 160 pixels. Our experimental results show that the DenseNet121 model achieved the highest accuracy of 98.68 % and AUC of 98.58% on the 160-pixels sub-database. These results suggest that our proposed work can assist pathologists in the detection of gastric cancer through histopathological image analysis.