Advanced Computational Intelligence Techniques Based Computer Aided Diagnosis System for Cervical Cancer Detection Using Pap Smear Images

Cervical cancer is the most well-known type of malignancy which exists among women around the world. According to an Indian Council of Medical Research report, nearly 100,000 women succumb to cervical cancer in India annually. Manual detection of cervical cancer becomes less effective due to subjective analysis, labor-intensive methods and time consumption. Hence, there arises the need for an automated system for cervical cancer detection. Even though many modalities such as the human papilloma virus (HPV) test, colposcopy, computerized tomography (CT) and magnetic resonance imaging (MRI) scan are available, the pap smear is the primary screening test which identifies abnormal cell changes. The proposed method uses two levels of classification using the deep learning technique, followed by a support vector machine (SVM) to tackle the overlapping cell issues. Using deep learning technique, the image patches are classified into three groups: nucleus, cytoplasm and background. The SVM classifier is then used to indicate the patient status as normal or abnormal. A supervised deep learning network is trained with 40% of nucleus, 40% of cytoplasm and 20% of background information. The new data are correctly classified into three groups using this trained deep learning network. Further, shape-based features are extracted from the nucleus region and are fed into the SVM to classify the cervical cells as normal or abnormal.The proposed system has achieved an accuracy of 97.5 and 90.9% for the deep learning technique and SVM classifier, respectively.