Multifeature Fusion Classification Method for Adaptive Endoscopic Ultrasonography Tumor Image

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Gastroenterology

Background:

• Endoscopic ultrasonography (EUS) aids in diagnosing digestive tract submucosal tumors.
• Subjective interpretation of EUS images leads to diagnostic variability.
• Computer-aided diagnosis systems can enhance accuracy and efficiency.

Purpose of the Study:

• To develop a multifeature fusion classification method for adaptive EUS tumor images.
• To improve the accuracy of computer-aided diagnosis for EUS images.
• To address the limitations of subjective interpretation in EUS diagnosis.

Main Methods:

• Region of interest selection based on prior information.
• Image gray histogram feature extraction with Principal Component Analysis (PCA) for dimensionality reduction.
• Fusion of reduced grayscale features with Local Binary Pattern (LBP) and Gray-Level Co-occurrence Matrix (GLCM) features.
• Classification using a multiclass support vector machine (SVM).

Main Results:

• The proposed multifeature fusion method achieved the highest accuracy of 96.18%.
• The area under the curve (AUC) reached 99%, indicating excellent performance.
• The method effectively learned gray distribution and combined multiple feature types.

Conclusions:

• The developed method significantly improves the classification accuracy of EUS tumor images.
• This computer-aided approach offers a valuable tool for diagnosing digestive tract submucosal tumors.
• The multifeature fusion strategy enhances diagnostic efficiency and overcomes subjective interpretation limitations.