Enhanced Cervical Cancer Diagnosis Using Advanced Transfer Learning Techniques
Gunjan Shandilya, Vatsala Anand, Rahul Singh Chauhan, Hemant Singh Pokhariya, Sheifali Gupta · 2024
Cervical cancer poses a significant worldwide health obstacle for women, and timely identification is essential for effective treatment. Pap smear tests are a cost-effective and efficient screening method. However, the implementation of automated recognition and categorization of Pap smear cells can greatly improve the early detection of abnormalities. This research utilizes a transfer learning methodology, employing the fine-tuned Xception pre-trained model to identify and classify cervical cancer images with a training accuracy of 96.23% and a validation accuracy of 95.56%. The study also incorporates the Xception architecture, which leads to a decrease in processing resources while still achieving a high level of validation accuracy. The proposed framework underwent training and testing on the SipakMed dataset comprising 4049 high-resolution images. It achieved an accuracy of 95.70% in accurately differentiating between five categories of cervical cell samples. Utilizing transfer learning with the Xception model yields notable enhancements in both efficiency and accuracy, presenting a hopeful resolution for diagnosing cervical cancer, especially in situations with limited resources. This automated Pap smear cell classification development contributes to worldwide initiatives aimed at promoting women's health by improving early detection and perhaps saving lives through prompt intervention.