Classification of Liver Diseases using CT and MR Images

Abstract

Over the last few decades, liver cancerous diseases have become a major leading cancer disease worldwide. The classes of liver cancer diseases enable further investigation with the diagnosis as the pre treatment. Early detection and diagnosis lessen the need for many implanted diagnostics, and receiving treatment sooner are advantageous. As a result, the current study was carried out to diagnose the many types of liver cancer diseases and their classifications, levels of tumor, and interpretation. The diagnosis of liver cancerous disorders is based on detecting and classifying the diseases: Hepatocellular carcinoma (HCC) and Metastases (MET). However, variability within the liver tumor portion allows a different class of intensities as the normal region and tumor region intensities mix-up. As a result, it is essential to perform the processing work that results in a varied tumor appearance in the images. An effective Computer Aided Diagnosis (CAD) system is designed to identify, classify, and interpret the level of liver tumor classes. We created a composite database that allows us to execute image processing work rapidly to complete the procedure. This database has a huge number of images for all tumor classes. The data images enable thorough analysis with all objectives work, resulting in the desired effect outcomes. In this research work, a composite database was created that included 4566 images from the CT and MR imaging domains for both the normal liver and the tumor region of the liver. Several classes in this composite database are described by the medical domain, including tumor classes and normal liver classes. There are 1957 abnormal CT and MR tumor images and 2609 normal CT and MR images. There are a total of 1054 abnormal MR tumor images and 2274 normal MR domain images. There are 844 abnormal CT tumor images and 395 normal CT domain images in total. For the tumor images, 600 MR images of HCC images and 513 MR images of MET were used, 361 CT images of HCC and 483 CT images of MET were used. These images were formed as a result of data collecting from several hospitals between January 2013 and July 2016. For the CT imaging, data were obtained from 15 patients, 15 normal liver patients, 4 HCC patients, and 16 MET patients. Four normal liver patients, six HCC patients, and nine MET patients were selected for MR imaging. Many of these patients have multi-phase data, meaning they have more than one phase of the same condition with varying specifications. The entire database, which has various aims, qualities, and levels of completeness. As a result, a data structure was created to handle the image processing work. Medical image factors include blur and artifacts that reduce contrast resolution, blurring two separate classes of intensities, and not allowing the required information to be obtained. Their presence complicates liver cancer diagnosis and interpretation. As a result, normal and malignant images of the liver are pre-processed. This is done utilizing the contrast limited adaptive histogram equalization (CLAHE) and constrained variational histogram equalization (CVHE) algorithms. Both approaches equalize histograms. Histogram equalization (HE) is image enhancement for low contrast images. As part of the pre-processing, photos are enhanced. Bilinear interpolation removes the CLAHE-induced boundary of these locations. CLAHE's CT image processing is robust, reliable, and flexible. With the CVHE method, you get the vi benefits of the histogram equalization algorithm in contrast enhancement for gray level images while preserving the overall quality of the image. A differentiated look with the tumor for varied class intensities resulted in the tumor increased regions. The improved images were verified and confirmed by qualified radiologists using visualization. An automated CAD technique for the detection and the classification of multi class liver cancer diseases using the CT and MR images is developed. The primary liver cancer is HCC, and the secondary liver cancer is MET. The CLAHE enhanced images offer a better tumor look on visualization than the original. The features are extracted from the enhanced images for the identification and categorization of liver tumors. For non-tumor types, the Region of Interest (ROI) may be the normal liver portion and for malignant types, the tumors region ROI can be 30 × 30 or 25 x 20 pixels. With these pixels, the class features are evaluated. First order statistics (FOS), Gray Tone Difference Matrix (GTDM), Gray Level Co Occurrence Matrix (GLCM), and Gray Level Run Length (GLRLM). This study employs numerous internal ROIs (ranging from 1-5) from a given enhanced image in the FOS, GLCM, and GTDM. Individual and group features are used to classify the IROI. The ratio feature compares the IROI and SROI feature values. Feature class may be a type of feature or a set of characteristics selected based on their individual classifier rate. A Genetic Algorithm (GA) is used to choose the IROI and ratio features automatically. SVM classifier classifies automatic and manual feature selection. With all the derived features utilizing ROI, tumor detection is possible using normal and malignant images. When a tumor is detected, it is classified as HCC or MET. Decision tree, Adaboost, Random Forest (RF), Support vector machine (SVM), Generalized Linear Model (GLM) and Neural network (NN) were used in study 2. Detection and classifier accuracy rate and area under the curve are improved using a multi-level ensemble (AUC). The sensitivity/specificity and recall/precision curves are assessed. The classifier rate was tested using k-fold cross-validation (CV).For the study 3, The Gabor filter scaling and orientation parameters are designed for the detection and classification. The optimized Gabor filter is used by five different scaling and eight orientation parameters, yields 40 different filters that cover the frequency domain, in multi resolution. The experiments are designed for detection and classification using CT, MR and the combination of the CT and MR images. The ROI is used as the inputs for the scaling and the orientation of Gabor filters using the classifier accuracy, AUC, sensitivity/specificity and the precision by all the six classifiers. The CV has been done on the best classifier to check their robustness with the detection and the classification experimentations. CT and MR images are segmented using the OKK-means clustering algorithm, which outperforms the previous accuracy, sensitivity, and specificity approaches. It eliminated noise with adaptive median filtering. Histogram equalization reduces the noise. The OKK-means clustering technique and the oppositional firefly algorithm (OFA) are used to choose the best features from the feature sets. Image registration is performed using a multi-phase CT image of the same patient. The moving and fixed images are selected to get the registered image for mutual information (MI). There are two extracted features: maximally stable extreme regions (MSER) and Speeded up robust features (SURF). vii This helps to give correct registered images. The registered images differ in color combinations. A good output form, color combination, output look, or proper difference of images offers MI with registered images. These results aid the radiologist as a second opinion for liver cancer diagnosis. The output parameters assist in the assessment of the tumor size and comment on tumor level in those images. Before treatment, the images of the tumor should be analyzed. The treatment should be based on the size and location of the liver tumor. This allows the radiologist to obtain tumor classes, show level information, and analyze many features of the tumor prior to diagnosis, utilizing both CT and MR images. To assist the radiologist, the more delayed phases contain more regions in the registered images. To gather more information, the radiologist needs to capture more stages images. Finally, the current research includes (i) liver image enhancement utilizing HEs methods, (ii) liver tumor identification and classification using ROIs and classifiers, (iii) segmentation of liver tumor, and (iv) registration of multi-phase CT images. Trying to assist radiologists, these computer-aided diagnosis experiments work provides the generalization of a clinical database, to analyze the position and size of the tumor and proves generalization ability, in real-time diagnosis of detection, classification, level of the tumor and the way to do the analysis before the treatment.