Abstract
Fuzzy enhancement is applied in computer-aided diagnosis (CAD) of liver cancer from B-mode ultrasound images as a pre-processing procedure. This method was evaluated using three classifiers—K-means, back-propagation neural network (BPNN), and support vector machine (SVM)—across 25 features derived from first-order statistics (FOS), gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level dependent matrix (GLDM), and LAWS texture energy. Analyzing 166 normal liver tissues, 30 hemangiomas, and 60 malignant tumors, the study demonstrates that fuzzy enhancement significantly improves classification accuracy across all models, proving it to be an efficient preprocessing procedure for CAD systems.
Introduction
Liver cancer remains a critical health challenge, particularly in Asian populations. While needle biopsies are the diagnostic "gold standard," they are invasive and susceptible to sampling errors. B-mode ultrasound is the preferred frontline tool due to its non-invasive and cost-effective nature; however, its diagnostic utility is often hampered by subjective image quality and speckle noise. This paper addresses these limitations by introducing a fuzzy logic-based preprocessing step designed to standardize and sharpen ultrasound images before automated classification.
Methodology
The proposed diagnostic pipeline moves from raw image acquisition to final classification through several distinct stages.
Fuzzy Enhancement
The core innovation is the application of fuzzy enhancement. Rather than traditional contrast stretching, the image is treated as a "fuzzy set." Each pixel is mapped to a "fuzzy property plane" where its value represents its membership degree to the maximum intensity level. A contrast intensification operator is then applied to this plane to "push" ambiguous pixel values toward bright or dark extremes, effectively sharpening the texture features essential for diagnosis.
Feature Extraction and Classification
Following enhancement, the system extracts 25 texture features, including basic first-order statistics and complex relationship matrices (GLCM, GLRLM, GLDM). These features are then processed by three machine learning models: K-means clustering, BP Neural Networks, and Support Vector Machines (SVM).
Key Results and Discussion
The study utilized a dataset of 256 ultrasound sonograms. The visual impact of the fuzzy enhancement is evident in the sharpened texture patterns of normal, benign, and malignant tissues.
Quantitative analysis revealed significant performance gains across all classifiers, with SVM emerging as the most effective:
- Malignant Tumor Detection (SVM): Accuracy rose from 96.67% to 100% after fuzzy enhancement.
- Benign Tumor Detection (SVM): Accuracy for identifying hemangiomas improved from 80% to 93.33%.
- Neural Network Improvement: The BPNN saw its sensitivity for liver cancer increase from 93.33% to 96.67%.
These results suggest that by operating in the fuzzy property plane, the algorithm can intelligently distinguish meaningful texture variations from random machine noise, providing cleaner data for the classifiers to process.
Conclusion
Fuzzy enhancement is a robust preprocessing tool for the computer-aided diagnosis of liver tumors. By reducing human subjectivity and machine-induced noise, it enables high-accuracy classification of liver pathologies. Future research will focus on automating the Region of Interest (ROI) selection to create a fully autonomous diagnostic system.
Summary updated