Utilizing metaheuristic optimization with transfer learning for efficient colorectal carcinoma detection in biomedical imaging

Abstract

Colorectal cancer (CRC) is the third most popular cancer across the world. Its morbidity and death are reduced by early screening and detection. The screening outcomes are enhanced by computer-aided detection (CAD) and artificial intelligence (AI) in screening models. Contemporary imaging technologies such as near-infrared (NIR) fluorescence and optical coherence tomography (OCT) are implemented to identify the early-phase CRC of the gastrointestinal tract (GI tract) via the identification of morphological and microvasculature changes. Most recently, deep learning (DL)-based approaches have been used directly on raw data. Nevertheless, they are hampered by biomedical data deficiency. These studies can enhance metaheuristic optimization using the transfer learning to detect colorectal cancer successfully (MHOTL-ECRCD). The MHOTL-ECRCD method concentrates on biomedical imaging of CRC categorization and detection. MHOTL-ECRCD minimizes noise through the process of adaptive bilateral filtering (ABF). In MHOTL-ECRCD methodology, Inception-ResNet-V2 is adopted to learn the inherent and complicated image preprocessing features thus used during feature extraction. To classify CRC and detect it, the gated recurrent unit (GRU) approach is applied. Lastly, parameters of the GRU model are optimized with a human evolutionary algorithm. Good classification results of MHOTL-ECRCD are demonstrated by a number of benchmark dataset trials. MHOTL-ECRCD technology superseded the recent techniques as large volumes of comparison were made.