Performance Analysis of Cervical Cancer Detection System Using Fusion Based CFICNN Classifier

This paper proposes a fully computer assisted automated cervical cancer detection method using cervical images.This proposed system consist of six modules as Edge detector, image fusion, Gabor transform, feature computation, classification algorithm and segmentation method.The edge pixels show the contrast edge variations of each pixel in cervical image with respect to its corresponding nearby pixels.Hence, these edge pixels are detected using fuzzy logic and then the edge detected cervical images are fused using arithmetic pixel fusion algorithm.This fused cervical image having the pixels in the form of spatial resolution and hence it is need to be converted into multi-format resolution for computing the features from it.The spatial pixels in fused image are converted into multi orientation pixels using Gabor transform and then features are computed from this Gabor image.In this work, Local Binary Pattern (LBP), Grey Level Co-occurrence Matrix (GLCM) and Pixel Intensity Features (PIF) are computed from the Gabor cervical image.These features have been classified by the Cervical Features Incorporated Convolutional Neural Networks (CFICNN) classification algorithm.The modified version of the Visual Geometry Group-Convolutional Neural Networks (VGG-CNN) architecture is called as Cervical Features Incorporated CNN (CFICNN) and it is proposed in this paper for both training and classification process.Finally, the cancer pixels are segmented using morphological operations based segmentation algorithm.The Guanacaste Dataset (GD) and Kaggle Dataset (KD) are used for estimating performance efficiency.