Screening of Cervical Cancer by Artificial Intelligence based Analysis of Digitized Papanicolaou-Smear Images

Introduction: Cervical cancer is one of the main causes of female mortality due to cancer in the world. Researchers have suggested that by regular cervical cancer screening its incidence and mortality a can be reduced by about 84%. Study aimed to apply multiple artificial intelligence based algorithms on the task of screening of cervical cancer and compare their working efficiency and to select the best performing algorithm and use that for developing intelligent software tool that can aid in the prognosis of cervical cancer. Material and methods: Fifteen different machine learning algorithms were implemented for screening of cervical cancer by analysis of digitized Papanicolaou-smear images using a primary dataset for training. Slides of Pap smear test of about of about 200 clinical cases were collected from multiple health care institutions in north India and images were captured using a digital microscope. Images of 8091 cervical cells were obtained from these slides, which were quantitatively profiled and calibrated to prepare a primary database that could reflect the morphological features of the cells of cervix. All the algorithms were trained using this database and were tested using a subset of unseen cells. The generalizing ability of algorithms was tested and its performance was evaluated against the human pathologist using percentage of correctly classified cells and sensitively. Results: Almost all the algorithms performed well for identifying the cells infected by cancer by classifying the cells according Bethesda system of classification. Among all the algorithms used, multiple backpropagation neural networks presented an efficiency of about 78.0%, whereas the efficiency of all other algorithms was in the range of 69% to 76%. Conclusion: The results demonstrate that the artificial intelligence based techniques can effectively be used for developing tools for mass level screening of cervical cancer.