Classification of normal and abnormal overlapped squamous cells in pap smear image

Cervical cancer is considered the second major cancer and a deadly disease found in the uterine cervix of sexually active women which is possible to treat if found at an early stage using a pre-screening procedure. One such procedure is Pap smear test which helps to find abnormal cervical cells which may lead to cancer. The manual analysis of these Pap smear cell samples is challenging and is even more complex when cells are overlapped which demands the need for an automated system to reduce the complexity. The software-based automated system can be implemented using image processing techniques which can be installed on any computer, making it easy to use, and low cost. The main objective of this paper is to classify the overlapped cells which are pre-segmented using Mid-point segmentation algorithm. The Convolutional Neural Network (CNN) model with Rectified Linear Unit (ReLU) classifier classifies the given input images into two classes - normal and abnormal. The paper concentrates on the performance comparison between proposed method to other works and also with the manual prediction done by a cytotechnician. The model uses 917 pap smear cell images from Herlev Dataset for training and testing. The proposed model is evaluated on performance measures like precision, recall, F-score, support. The results reflect that the proposed model is best suited for Pap smear test analysis with 96% accuracy. Hence, proposed work makes a useful assistive tool for radiologists and clinicians to detect cervical cell abnormalities from pap smear cytology images.