classification and prediction of ovarian cysts using machine and deep learning techniques

Abstract

The research advancements in the field of image processing enable us to quantitatively analyse and visualize all modalities of medical images such as X-ray, Computerized Tomography (CT), Magnetic Resonance Imaging (MRI) and ultrasonic images. Conceiving the four modalities, Ultra sound images diagnosis is communally used unless the abnormalities are complicated. In such case the MRI may be needed for further diagnosis and surgery planning. newlineThis research utilizes medical images taken with due permission from the private scan centres such as SONA scan centre and Pixel scan centre in Trichy and from Indra Clinic, Nagerciol. Also some of the online images are collected from the Kaggles data source. Throughout this work, care has been taken to ensure that the database employed contains images representing different ages, and pathologies. newlineIn this research, six different classes of Ultra sound images are taken, namely malignant cyst, simple cyst, Polycystic Ovarian Syndrome (PCOS), dermoid cyst, haemorrhagic cyst and endrometriotic cyst. The proposed work process involves pre-processing of ultrasound ovarian cyst images to make them fit for further processing. It is followed by segmentation, feature extraction and classification of ovarian cyst types. Initially, the ultrasound scan images are collected and pre-processed using M3 filter and two conversion method i.e., image conversion from BGRand#8594;RGB and RGBand#8594;HSV to remove the unwanted noise and to enhance the ovarian cyst images. Then, the feature extraction process is performed using Hu Moment, Haralick texture and color histogram to extract the required features from pre-processed cyst images. Moreover, segmentation process is performed using bitwise operator for machine learning model and cluster, region and edge-based segmentation method for deep learning neural network. Finally, the classification process is newlineiv newlinecarried out using different ML and DL methods and the performance of the models were evaluated. newline