Enhancing Cervical Cancer Detection: Leveraging Semi-Supervised Learning for Improved Prognosis

Cervical cancer persists as a life-threatening issue among women worldwide. Early-stage detection of cancer leads to better survival rates and lower treatment costs, thereby impacting the patient’s overall betterment. Therefore, cervical cancer detection and appropriate prognosis treatment in the early stages are crucial. For diagnosing cervical cancer, patients undergo tests and scanning like, a Pap test or Pap smear screening, where a considerable amount of data samples are required. To mitigate the need for a large number of labeled images, we used a semi-supervised learning framework, named FixMatch, which classified various types of cervical cancer with a smaller number of labeled images. The semi-supervised FixMatch framework worked with 30%, 40%, 50%, and 60% of the labeled data, where the 50% resulted in a notable accuracy of 94%, which is close to three supervised models( ResNet 101, DenseNet169, and EfficientNetB4). This emphasizes the importance of the labeled data ratio variation on semi-supervised model effectiveness. Along with the critical findings by varying data proportion and reduced reliance on extensive labeled data, the integrated model offers noteworthy generalizability to reduce cancer fatality. The model will be able to aid in global healthcare campaigns and increase crucial medical progress among multiple medical domains.