Efficient gastric tumor detection from endoscopic images using trans-mapped learning models

Gastric cancer has emerged as a major health concern in recent years, often attributed to improper or unhealthy dietary habits. Early detection remains challenging due to the lack of identifiable symptoms in its initial stages, emphasizing the need for intelligent computational diagnostic methods. This study introduces the Inflate Region-based Tumor Recognition (IRTR) scheme, a novel approach leveraging endoscopy images and trans-mapped learning to detect inflated tumor regions with precision. The proposed scheme employs trans-mapping layers, which are trained to analyze inputs and outputs for identifying high and low-intensity feature regions. By focusing on external boundaries with elevated trans-intensity levels, the scheme effectively identifies regions exhibiting significant differences across the image. These mapped features are then utilized to train a model that repetitively processes high-to-low and low-to-high intensity transitions across input and output layers, enhancing the recognition of inflated tumor regions. Boundary differentiation, a key component of this approach, further refines detection precision from early endoscopic inputs. Evaluation results demonstrate that the IRTR scheme achieves superior performance, with an accuracy improvement of 9.38%, a precision increase of 12.04%, 9.69% in specificity and a mean error reduction of 11.04% for maximum intensity rates. This study underscores the potential of trans-mapped learning in advancing early gastric tumor detection.