Gastrointestinal Cancer Detection and Classification Using African Vulture Optimization Algorithm With Transfer Learning

Abstract

Gastrointestinal (GI) cancer comprises esophageal, gastric, colon and rectal tumors. The diagnosis of GI cancer often relies on medical imaging modalities namely magnetic resonance imaging (MRI), histopathological slides, endoscopy, and computed tomography (CT) scans. This provides particular details about the size, location, and characteristics of tumors. The high death rate for GI cancer patients shows that it is possible to increase analysis for a more personalized therapy strategy which leads to improved prognosis and few side effects although many extrapolative and predictive biomarkers exist. Gastrointestinal cancer classification is a challenging but vital area of research and application within medical imaging and machine learning. Artificial intelligence (AI) based diagnostic support system, especially convolution neural network (CNN) based image examination tool, has enormous potential in medical computer vision. The study presents a GI Cancer Detection and Classification utilizing the African Vulture Optimization Algorithm with Transfer Learning (GICDC-AVOADL) methodology. The major aim of the GICDC-AVOADL model is to examine GI images for the identification of cancer. To achieve this, the GICDC-AVOADL method makes use of an improved EfficientNet-B5 method to learn features from input images. Furthermore, AVOA is exploited for optimum hyperparameter selection of the improved EfficientNet-B5 method. The GICDC-AVOADL technique applies dilated convolutional autoencoder (DCAE) For GI cancer detection and classification. A complete set of simulations was conducted to examine the enhanced GI cancer detection performance of the GICDC-AVOADL technique. The extensive results inferred superior performance of the GICDC-AVOADL algorithm over current models.