<![CDATA[Early detection of gastric pathologies, such as polyps, esophagitis, and ulcerative colitis, plays a pivotal role in improving patient clinical outcomes and long-term treatment efficacy. Despite advancements in medical imaging, manual endoscopic analysis remains a labor-intensive process prone to human error and inter-observer variability, creating a critical research gap for automated diagnostic tools. This research introduces a robust automated classification framework employing the ResNet18 architecture, optimized through a refined Transfer Learning methodology. The study utilizes a comprehensive multi-class dataset, with input data undergoing meticulous preprocessing, including global normalization and strategic data augmentation, to enhance generalization. Empirical evaluations conducted over 50 epochs revealed superior performance, with the proposed model achieving an overall accuracy of 94.05%. Notably, a precision rate of 100% was attained, indicating zero false alarms, while a high sensitivity of 91.67% confirmed the model's effectiveness in distinguishing subtle cancerous features from healthy gastric folds. These quantitative findings underscore the framework's reliability and its potential for seamless integration into clinical decision-support systems. By providing high-fidelity diagnostic assistance, this study contributes to the evolution of computer-aided diagnosis (CAD), offering a scalable solution to reduce clinician workload while significantly increasing the accuracy of early-stage gastric pathology detection.]]>
