G.B. SOLOMON*#, A.G. ABEBE*, D.S. TEFERI**, M.M. MENGISTE***
*Department of Physics, College of Natural and Computational Sciences, Haramaya University, Haramaya, Ethiopia, #e-mail: binysoul@gmail.com
**Radiology Department, College of Health Sciences, Tikur Anbessa Specialized Hospital, Addis Ababa University, Addis Ababa, Ethiopia
***Department of Medicine, College of Health Science and School of Medicine, Dire Dawa University, Dire Dawa, Ethiopia
The aim of this research was to develop an appropriate algorithm that can automatically detect and classify liver cancers. Digital image processing (DIP) provides different techniques for the segmentation and classification purposes. The sample was taken from 36 abdominal CT scans from Tikur Anbessa Specialized Hospital in DICOM stored in Addis Ababa. Semi-automatic techniques were used to segment liver images from the abdominal CT images then artificial neural network (ANN) was applied for classification of liver out of the given sample through nine texture features into 12 normal and 24 abnormal images. Then, liver tumors were segmented from the 24 abnormal liver images using the histogram-based Otsu method. The classification of liver tumors was performed by using ANN through six morphological features into 12 benign and 12 malignant. The classification results are presented in this paper using the confusion matrix, which shows 95.8 % maximum accuracy rate of tumor classification. However, the performance of the classifier could be improved by more sample images. The outcome of this work may help radiologists to identify tumors at early stages and to classify them as benign or malignant. The methodology proposed here might reduce the fatality rate of liver cancers in Ethiopia.
Key words: Digital image processing, ANN, liver cancers, CT images.
Corresponding author’s e-mail: binysoul@gmail.com