Background: Liver cancer due to hepatic tumors is one of the primary mortality causes globally. Detecting and diagnosing such tumors can be very tricky in reducing death rates. Segmenting liver tumors from computed tomography (CT) images is a very tricky and challenging task due to many factors such as fuzziness of the liver intensities range, liver pixels intensities values intersection with the neighboring abdomen organs, noise imposed by CT scanner, and variance in tumors and appearances. This paper introduces a Computer-Aided Diagnosis (CAD) system based on some of the Fuzzy C-means (FCM) method variations to detect liver tumors on CT images. Furthermore, one of the main objectives of this paper is to diagnose and label detected liver tumors, either benign or malignant.
Methods: Multi-scale Fuzzy C-Means (MSFCM) is used as the liver segmentation method, and its output is the liver segmented out of the abdomen CT image. In order to achieve high-quality liver tumors segmentation, the segmentation is done using two FCM algorithms; Gaussian Kernelized Fuzzy C-Means (GKFCM) and Fast Generalized Fuzzy C-Means (FGFCM), respectively. The diagnosis is achieved by extracting features from segmented tumors and passing them to the support vector machines (SVM) classifier.
Results: To evaluate the overall performance of the proposed CAD system, the system was implemented using CT images from three liver benchmark datasets with a total of 250 subjects. The used datasets are MICCAI-Sliver07, LiTS17 and 3Dircadb. Different performance metrics were calculated, such as accuracy (ACC), sensitivity (SEN), specificity (SPE), and dice similarity score (DSC). The proposed system achieved reasonable performance results with an average ACC, SEN, SPE, and DSC of 96.62%, 95.84%,94.20%, and 95.21%, respectively.
Conclusions: The proposed system was able to differentiate benign and malignant tumors with reasonable accuracy. The resulted accuracy resulted from our CAD system features such as noise robustness, detail persevering, and being a fully-automatic system with no need for user interaction. The experimental results showed the applicability of the proposed system using different liver datasets.