Hyperparameter Optimisation for Breast Cancer Detection Using APO and Pre-Trained CNNs

Abstract

Early detection of breast cancer improves survival rates and treatment outcomes. Mammography remains the key diagnostic technique; however, building deep learning models for reliable categorisation is challenging. This paper presents a groundbreaking method for fine-tuning hyperparameters in two cutting-edge convolutional neural networks (CNNs), ConvNeXtBase and ResNet-50, which employ the Arctic Puffin Optimisation (APO) algorithm. Experiments were performed on two benchmark mammography datasets: CBIS-DDSM and MIAS. The APO-optimised ConvNeXtBase model achieved 98.46% accuracy on the CBIS-DDSM dataset and 99.34% on the MIAS dataset, with precision and recall both at 100% in the latter. These findings indicate that APO increases CNN performance, making it a promising tool for computer-assisted breast cancer diagnosis.