Key points are not available for this paper at this time.
One of the leading and most serious causes of death worldwide is liver tumours. Currently, one of the most well-known Computer Vision disciplines (CV) is Medical Imaging (MI), which aids radiologists and physicians in the early detection and diagnosis of liver tumours. In order to diagnose patients, radiologists and doctors read hundreds of images using manual or semi-automated technologies, such as Magnetic resonance imaging (MRI). Consequently, a fully automated approach is required to use MRI scans, the most common and extensively used imaging modality, for early tumour detection and diagnosis. Implementing machine learning (ML) techniques in finding the tumour affected parts are the main focus of the proposed work. The multiclass liver tumour classification procedure followed in the tumour identification process follows, Random Forest (RF), Random Tree (RT), Logistic Model Tree (LMT) and J48 with multiple automated Region of Interest (ROI). Hemangioma, cyst, hepatocellular carcinoma, and metastasis are the four basic tumour classes, which are to be classified before severity increases. Transformation procedures followed in image processing are conversion of collected MRI images into grayscale, then boosting the contrast with histogram equations. The Gabor filter was used to minimise noise, and an image sharpening technique was used to enhance image quality. Furthermore, employing texture, binary, histogram, and rotational, scalability, and translational (RST) algorithms, 55 characteristics of various pixel dimensions were obtained for each ROI. Out of these 55 characteristics, 20 optimised features were chosen for classification using the correlated feature selection approach. RF and RT outperformed J48 and LMT, with respective accuracy of 97.48% and 97.08%, according to the collected data. The new framework that has been suggested will aid doctors and radiologists in the diagnosis of liver tumours.
Bhagya et al. (Fri,) studied this question.