Deep Learning for Cervical Cancer Classification: Leveraging Transfer Learning for Multiclass Diagnosis
Gunjan Shandilya, Vatsala Anand, Rahul Singh Chauhan, Hemant Singh Pokhariya, Sheifali Gupta · 2024
Annually, more than 500,000 new cases of cervical cancer are recorded worldwide, posing a substantial health risk to women. Although Pap smears and HPV tests are currently available for screening, there is a significant requirement for more advanced procedures to promote early detection and diagnosis. The current body of research emphasizes the capability of Convolutional Neural Networks (CNNs) in classifying medical images. However, there is a lack of studies that explore the use of advanced architectures, such as InceptionV3, specifically for detecting cervical cancer. This study aims to fill this void by examining the application of InceptionV3 in a convolutional neural network (CNN) structure for transfer learning to classify cervical cancer based on clinical photos automatically. The primary achievement of this research is the creation and verification of an exceptionally precise model, attaining a 94.89% accuracy in distinguishing distinct characteristics among five categories of cervical cell samples from the SipakMed dataset, comprising of 4049 high-resolution images. This innovation resides in the utilization of InceptionV3, which not only reduces the amount of processing resources required but also surpasses earlier neural architectures in terms of validation accuracy. This breakthrough offers a hopeful resolution for widespread cervical cancer diagnosis, specifically in regions with limited resources. This study substantially contributes to global efforts to enhance women's health by addressing the existing research gap and improving early detection approaches.