Fusion of Texture Features Applied to H. pylori Infection Classification from Histopathological Images.

Area of Science:

• Medical Diagnostics
• Computational Pathology
• Bacteriology

Background:

• Helicobacter pylori (H. pylori) infection affects billions globally, causing significant gastric diseases.
• Current diagnosis relies on manual histopathological analysis of biopsies, which is labor-intensive, time-consuming, and prone to errors.
• The cost and accessibility of traditional H. pylori diagnosis can be limiting factors.

Purpose of the Study:

• To evaluate the efficacy of texture features for automated binary classification of H. pylori in histopathological images.
• To investigate the impact of combining texture features using Particle Swarm Optimization (PSO) for improved classification performance.
• To compare the performance of texture analysis methods against state-of-the-art deep learning techniques.

Main Methods:

• Extracted various texture features from 204 histopathological images, excluding color information.
• Employed Particle Swarm Optimization (PSO) to optimize the combination of selected texture features.
• Performed binary classification to differentiate H. pylori-positive from H. pylori-negative cases.

Main Results:

• Texture analysis methods achieved a high accuracy of 94.61% and an F1-score of 94.47%.
• The optimized combination of texture features demonstrated a robust balance between precision and recall.
• The proposed method surpassed the performance of ResNet-101 by 4.41%.

Conclusions:

• Traditional texture analysis methods remain highly competitive for H. pylori detection in histopathology.
• Automated texture feature analysis offers a promising, efficient, and accurate approach for H. pylori diagnosis.
• This technique has the potential to improve diagnostic speed and reduce costs compared to manual methods.