Abstract

Liver cancer detection through automatic segmentation of both liver and tumour structures in abdominal CT images presents a significant challenge due to the complex nature of soft tissue variation in terms of intensity, shape, and location. Traditional methods relying on grey-level or shape-based techniques often fail to provide accurate results. This study proposes an advanced approach that integrates Gabor Features (GF) with three distinct machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), and Deep Neural Networks (DNN)—for more efficient liver and tumour segmentation. The GF-based texture data is designed to maintain consistency and homogeneity across different slices of the liver, improving the accuracy of segmentation. In the first phase, pixel-level features are extracted using a series of Gabor filters, capturing the intricate texture of the liver. Subsequently, three classifiers (RF, SVM, and DNN) are employed on the GF data to segment the liver from the CT image. In the final phase, tumour segmentation is performed on the liver region using the same features and classifiers. The Gabor filter's analogy to the human visual system enhances the method's robustness and performance, ensuring accurate pixel-wise segmentation. The proposed model demonstrates superior potential for liver cancer detection, offering a promising tool for clinical diagnosis and decision-making.