Variable Kernel Feature Fusion and Transfer Learning for Pap Smear Image-Based Cervical Cancer Classification
S. Priya, Vargheese Mary Amala Bai · International Journal of Electronics and Communication Engineering · 2024
Cervical cancer, a malignant tumour that forms in the cervix, significantly contributes to cancer-related mortality among women globally, making early diagnosis crucial for effective treatment. Pap smear images, which are microscopic images of cervical cells, are commonly used for the detection of abnormal cells that may lead to cervical cancer. This study introduces a novel classification approach, the Variable Kernel Feature Fusion-CNN (VKFF-CNN), which improves classification performance by fusing multi-scale features using convolutional layers with 3x3, 4x4, and 5x5 kernels. This architecture captures a diverse set of features, enhancing the ability of the model to accurately classify cervical cells. With an average accuracy of 98.03%, precision of 97.83%, recall of 97.11%, and an F1 score of 98.23%, the VKFF-CNN exhibited outstanding outcomes on the Herlev Pap Smear dataset. These results demonstrate that VKFF-CNN significantly outperforms traditional machine learning models. The model’s confusion matrix indicated fewer misclassifications, underscoring its robustness and effectiveness. Including batch normalization and the softmax activation function further enhanced the model’s stability and accurate classification. Overall, VKFF-CNN presents a promising advancement for automated cervical cancer screening, providing highly accurate and reliable detection.