Leveraging Nature-Inspired Search to Boost CNN Performance in Colorectal Cancer Classification

Cancer affecting the colorectal region (CRC) is a leading contributor to Worldwide cancer-associated mortality, where timely and precise diagnosis is critical for enhancing patient prognosis. Although histopathology remains the benchmark for diagnostic accuracy, its manual assessment is time-consuming and subject to variability among pathologists. To address these challenges, this paper proposes a novel optimization framework CrcMRFA based on the Manta Ray Foraging Optimization (MRFO) algorithm to fine-tune Convolutional neural nets initial-ized with learned weights from prior training for histopathological image classification. Three architectures—VGG16, ResNet50, and DenseNet121—were optimized with respect to key hy-perparameters, encompassing parameters such as learning rate, batch size, dropout rate, and the number of trainable layers. Experimental evaluation Leveraging the Kather_texture_2016_image_tiles_5000 dataset demonstrated significant performance en-hancements across all metrics. The optimized ResNet50 achieved the best results, with accuracy improving from 90.32% to 95.97% and the Weighted Sum Metric (WSM) exceeding 96.77%. These findings highlight the potential of MRFO in automating CNN optimization for robust and ef-ficient CRC tissue classification.