Analysis and Classification of Different Gynecological Tumors Using Machine Learning Techniques

Abstract

Cancer is an uncontrolled development of cells. Mortality rate can be reduced by early newlinedetection of the cancer. Gynecological cancer arises in female reproductive system. It newlineis currently the second-leading cause of cancer deaths among women. Uterus, Ovarian, newlineCervical, Vagina and Vulva are five major different gynecological tumors. Uterus and newlineOvarian tumors are the most common cancer in developed counties. Uterus cancer is newlinediagnosed by biopsy, whereas ovarian tumor is diagnosed usually in advanced stage newlineusing MRI (Magnetic Resonance Image) or CT (Computed Tomography) imaging newlinemodalities. Because of increasing death rate of ovarian tumor in developed counties, newlinethere is a need of computational analysis tool to diagnose. newlineThe core objective of this research work is to reduce the number of newlinemisclassifications of ovarian tumours while increasing diagnostic accuracy using newlinemachine learning approaches. The input image dataset for ovarian tumour is collected newlinefrom the publicly available source The Cancer Genome Atlas (TCGA) data portal. It newlinecontains clinical, genetic, and pathological data of different cancer patients. TCGA in newlineconnection with The Cancer Imaging Archive (TCIA) provides largest radiology data newlinerepository in terms of DICOM (Digital Imaging and Communications in Medicine) newlineformat. This file includes patient data and an image data. To separate the patient data newlineand image data in a DICOM file, it is pre-processed by converting into Tag Image File newlineFormat (TIFF) images. The pre-processed TIFF images are labelled as early stage and newlinemalignant classes. newlineDeep Convolutional Neural Network (DCNN) models are employed newlineextensively in medical image diagnosis. Initially, three Deep Neural Network models newlineXception, ResNet50V2 and ResNet50V2 with FPN are used to train pre-processed newlinedataset and then analysed the performance of above classifier models by two measures, newlineaccuracy and loss. Compared to other models.ResNet50V2 with FPN model reduces newlinethe number of misclassification rate. newlineTo further improve the ResNet50V2F