A computer vision approach for the classification of multi liver tumor using computed tomography scan

Area of Science:

• Medical Imaging
• Computer Vision
• Oncology

Background:

• Accurate liver tumor classification is crucial for effective treatment planning.
• Distinguishing between benign and malignant liver tumors from medical images presents challenges.
• Computer vision offers potential for automated analysis of radiological data.

Purpose of the Study:

• To evaluate the efficacy of computer vision techniques for classifying benign and malignant liver tumors using computed tomography (CT) images.
• To develop and assess a robust feature extraction and selection pipeline for liver tumor analysis.
• To compare the performance of various machine learning classifiers in liver tumor detection.

Main Methods:

• A dataset of 900 2D CT images representing six types of liver tumors was utilized.
• Image preprocessing involved noise reduction using a Mean filter and region of interest selection.
• Feature extraction generated 67 multi-features, with 21 optimized features selected using a correlation-based feature selection (CFS) algorithm.
• Six classifiers, including multilayer perceptron (MLP), were trained and evaluated using 10-fold cross-validation.

Main Results:

• The multilayer perceptron (MLP) classifier achieved the highest accuracy of 97.67% on the optimized feature dataset.
• The developed computer vision approach demonstrated systematic and resilient performance across different imaging standards.
• The study successfully identified key features for differentiating liver tumor types.

Conclusions:

• Computer vision, particularly using MLP, shows significant potential for accurate liver tumor classification from CT images.
• The proposed method can serve as a valuable tool to assist radiologists in diagnosing liver diseases.
• Optimized feature selection enhances the performance and reliability of automated diagnostic systems.