Efficient Cervical Cancer Segmentation and Classification Using Hybridized RBFNN Approach with Deep Learning Model

Abstract

newlineCervical cancer remains a significant health concern, impacting over half a million newlinewomen annually and ranking among the leading causes of female mortality. Early newlinedetection is crucial for effective treatment, yet the challenges posed by the costly and newlinelabor-intensive nature of cancer detection necessitate innovative solutions. Cervical newlinecancer classification involves the use of machine learning algorithms to analyze and newlinecategorize cervical tissue samples, typically obtained through various screening methods, newlineto identify whether the tissue exhibits signs of cancerous growth. This classification newlineprocess is crucial for early detection, which is vital for effective treatment and improved newlinepatient outcomes. newlineThis research proposes a comprehensive approach for cervical cancer detection utilizing newlinedata from the Kaggle Cervical Cancer Dataset, UCI Repository Cervical Cancer Data Set, newlineand Data World Cervical Cancer Data Set. The methodology involves an Improved newlineMedian Filter(IMF) for noise removal and a modified fuzzy GLCM-based segmentation newlinemethod to enhance data accuracy. Subsequently, radial basis function (RBF) neural newline newlinenetworks are employed for classification, and a novel hybrid particle swarm optimization newline fruit fly optimization algorithm (PSO-FOA) is developed to optimize RBF neural newlinenetwork weights, thereby improving model convergence. newlineIn the Kaggle dataset, the proposed Hybridized RBFNN (HRBFNN) model outperforms newlineothers with an accuracy of 94.75%, precision of 0.93, and recall of 0.92, showcasing its newlinesuperior performance. Similar trends are observed in the UCI Repository dataset and newlineData World dataset, emphasizing the consistent superiority of HRBFNN. This research newlinecontributes to advancing cervical cancer detection methodologies, offering a reliable and newlineefficient approach with implications for improving population-wide screening.