ColpoNet for automated cervical cancer screening using colposcopy images

Cervical cancer is one among the trivial forms of cancer that counts for 6.6% of all females cancers with an estimated 570,000 new cases in 2018. The mortality rate due to cervical cancer is approximately 90% in low or middle income countries due to lack of suitable pre-screening procedures and experienced medical staff. Colposcopy images or cervigrams, are the images that capture the cervical region, are considered as the gold standard by the medical experts for the identification and evaluation of cervical cancer. The visual assessment of cervigrams for recognizing cancer suffers from high inter- or intra-variations especially among less or unskilled medical experts. However, this method is dependent on colposcopists’ observation and it is more time consuming, tedious and laborious task which calls for development of computer-aided method for diagnosis of cervical cancer. With the technological advancements, deep learning has been commonly employed for providing automated solutions for disease diagnosis due to its self-learning capability. This paper presents a deep-learning-based method for cervix cancer classification using colposcopy images. The architecture of the proposed method namely, ColpoNet, has been motivated by the DenseNet model because it is computationally more efficient as compared to other models. Further, the method has been tested and validated on the dataset released by the National Cancer Institute and it has been compared with other deep-learning models namely AlexNet, VGG16, ResNet50, LeNet and GoogleNet to check scope of its applicability. The experimental analysis revealed that ColpoNet achieved an accuracy of 81.353% and shows the highest performance rate as compared to other state-of-the-art deep techniques. Such classification system can be deployed in clinics to enhance the early detection of cervical cancer in less developed countries.