Pap smear classification using combination of global significant value, texture statistical features and time series features

Pap smear test is routinely used today to diagnose cervical cancer. In the last 50 years, this simple test has saved millions of women's lives. Although successful, pap smears are not always perfectly analyzed and it is time-consuming work. Computer-assisted screening can be widely used for cervical cancer diagnosis today. So far, a variety of image processing based methods have been proposed for cervical cancer detection. Most of the previous methods usually addressed one of two challenges of accuracy or complexity. In most medical imaging laboratories the quality of the test images are low. Also, most of the pap smear classification methods are not resistant to rotation and gray-scale variations. Therefore, in this paper, the main motive is to provide a method that, along with the high diagnosis accuracy, has less computational complexity, and also is resistant to rotation and gray-scale variation. The proposed approach extracts textural and statistical information of the nucleus and cytoplasm of the cell image. Selected subset of harlick features, global significant value and time series features are employed in feature extraction phase. The proposed approach is evaluated using herlev dataset, which consist of 917 images in two normal and cancer classes. Our approach has been compared with some state-of-the-art methods. It is experimentally demonstrated that the proposed approach achieves the highest accuracy. Computational complexity is computed based on the running time and number of extracted dimensions. Low computational complexity and rotation invariant are some other advantages of the proposed approach.