Classifying Gastric Histopathology Images Using Hybrid Deep Feature Extraction and Vision Transformer Model

Gastric cancer symptoms are mild or not visible, which makes early detection challenging. The analysis of tissues obtained from the gastric lining depends on investigating the histopathology, but qualified pathologists must interpret these intricate images. However, due to the large volume of data, hectic routines of specialists, and fewer experienced personnel, it is quite time-consuming and challenging. In this study, an advanced computer-aided diagnostic (CAD) framework using histopathology images for early GC detection is proposed to tackle this issue. Deep features are extracted using the discrete wavelet transform, local binary pattern, fuzzy color histogram, and gray-level co-occurrence matrix. Furthermore, a Vision Transformer improves classification performance using its attention mechanism. The proposed hybrid CAD framework proved a trustworthy diagnostic tool, achieving 97% accuracy on a publicly available dataset.