Cervical cancer diagnosis based on modified uniform local ternary patterns and feed forward multilayer network optimized by genetic algorithm

Abstract

Cervical cancer is one of the most common types of cancer for women. Early and accurate diagnosis can save the patient's life. Pap smear testing is nowadays commonly used to diagnose cervical cancer. The type, structure and size of the cervical cells in pap smears images are major factors which are used by specialist doctors to diagnosis abnormality. Various image processing-based approaches have been proposed to acquire pap smear images and diagnose cervical cancer in pap smears images. Accuracy is usually the primary objective in evaluating the performance of these systems. In this paper, a two-stage method for pap smear image classification is presented. The aim of the first stage is to extract texture information of the cytoplasm and nucleolus jointly. For this purpose, the pap smear image is first segmented using the appropriate threshold. Then, a texture descriptor is proposed titled modified uniform local ternary patterns (MULTP), to describe the local textural features. Secondly, an optimized multi-layer feed-forward neural network is used to classify the pap smear images. The proposed deep neural network is optimized using genetic algorithm in terms of number of hidden layers and hidden nodes. In this respect, an innovative chromosome representation and cross-over process is proposed to handle these parameters. The performance of the proposed method is evaluated on the Herlev database and compared with many other efficient methods in this scope under the same validation conditions. The results show that the detection accuracy of the proposed method is higher than the compared methods. Insensitivity to image rotation is one of the major advantages of the proposed method. Results show that the proposed method has the capability to be used in online problems because of low run time. The proposed texture descriptor, MULTP is a general operator which can be used in many computer vision problems to describe texture properties of image. Also, the proposed optimization algorithm can be used in deep-networks to improve performance.