This manuscript introduces an automated Computer-Aided Classification (CAD) system to classify liver lesions into Benign or Malignant. The system consists of three stages; firstly, automatic liver segmentation and lesion detection. Secondly, extracting features from multiple ROIs, which is the novelty. Finally, classifying liver lesions into benign and malignant. The proposed system divides a segmented lesion into three areas, i.e. inside, outside and border areas. This is because the inside lesion, boundary, and surrounding lesion area contribute different information about the lesion. The features are extracted from the three areas and used to build a new feature vector to feed a classifier. The novelty lies in using the features from the multiple ROIs, and particularly the surrounding area (outside), because the malignant lesion affects the surrounding area differently compared to a benign lesion. Utilising the features from inside, border, and outside lesion area supports better differentiation between benign and malignant lesions. The experimental results showed an enhancement in the classification accuracy (using multiple ROI technique) compared to the accuracy using a single ROI.
|Title of host publication||Procedia Computer Science|
|Publication status||Published - 25 Jul 2016|
|Event||20th Conference on Medical Image Understanding and Analysis - Loughborough, United Kingdom|
Duration: 6 Jul 2016 → 8 Jul 2016
|Conference||20th Conference on Medical Image Understanding and Analysis|
|Period||6/07/16 → 8/07/16|