Advanced Detection of Gastrointestinal Cancer & Classification Using ASHMSO and Multi-Kernel Attention Densenet

Abstract

Gastric cancer presents as a prevalent and often fatal condition. Accurately differentiating between early-stage (EGC) and advanced-stage (AGC) gastric cancer is essential for developing customized treatment plans. Currently, computed tomography (CT) lacks the necessary diagnostic accuracy for gastric cancer staging to meet clinical needs. Numerous investigations resort to labor-intensive manual scrutiny of lesion sites, rendering it unfeasible for clinical assessment. Hence, this research proposes an advanced approach employing deep learning methodologies to identify and classify this particular cancer type. The methodology encompasses the following stages: a) Collection of a dataset comprising 192,312 images from a well-known repository, namely Kaggle. b) Preprocessing the raw images involves applying the Richardson median filter and contrast enhancement techniques. c) Feature extraction is performed using an Autoencoder. d) Feature Selection employs Advanced Snail Homing and Mating Search Optimization Algorithm (ASHMSO). e) Detection is accomplished utilizing the Multi-Kernel Merged Attention Densenet 169 classifier. Experimental investigations were carried out to assess the MKA-Densenet 169 system compared to several cutting-edge models across multiple metrics. The outcomes underscore the outstanding performance of the suggested system, attaining an accuracy rate of 0.979 along with a sensitivity 0.98 and specificity of 0.97 each.