Cervical cancer diagnosis using convolution neural network: feature learning and transfer learning approaches

Cervical Cancer most often affects women and is a feared disease worldwide. It occurs due to the abnormal growth of cells in the cervix and slowly spreads to other organs in the body. Cervical cancer can be cured completely if detected at an early stage. Various cervical screening methods are available, but early screening with low-cost Pap smear tests is the key to curing. Knowing about this cancer, if we contribute to detecting it at an early stage, may help to save a life. For accurate disease diagnosis and prompt care at an early stage of cell development, automated Pap smear cell recognition and classification are critical. Various computer-aided systems are available that use hand-crafted features of Pap smear images to categorize them into normal and abnormal cells. Deep learning classification methods have been the top choice as their approaches have attained state-of-the-art in tumor identification. In this article, deep learning is used in two different ways. In this approach, pre-trained models as feature extractors and different machine learning algorithms are utilized for classification; the second approach used is transfer learning using pre-trained models to classify Cervical Cancer images. Using the first method with a pre-trained model as a feature extractor, ResNet-50 achieves the highest classification accuracy of 92.03%. Using the second approach of fine-tuning pre-trained models, Google Net has given the highest classification accuracy of 96.01%.

Automated summary

Cervical cancer is a feared disease that primarily affects women worldwide. Early detection through low-cost Pap smear tests is crucial for complete cure. Automated recognition and classification of Pap smear cells using deep learning methods have been proven effective in accurate diagnosis and prompt care.