A non-entropy-based optimal multilevel threshold selection technique for COVID-19 X-ray images using chance-based...

Area of Science:

• Medical Image Analysis
• Computer Vision
• Artificial Intelligence

Background:

• Entropy-based image thresholding methods struggle with non-uniform grey value distribution in medical images.
• Accurate segmentation of COVID-19 X-ray images is crucial for research and diagnosis.
• Existing methods have limitations in handling complex image characteristics.

Purpose of the Study:

• To develop an efficient, non-entropy-based thresholding method for segmenting COVID-19 X-ray images.
• To introduce a novel fitness function (Segmentation Score) to minimize segmentation error.
• To enhance optimization techniques for improved segmentation accuracy.

Main Methods:

• A novel non-entropy-based thresholding method is proposed.
• A new fitness function, Segmentation Score (SS), is introduced to reduce segmentation error.
• A soft computing approach employing a chance-based birds' intelligence optimizer is utilized for maximizing fitness values.
• The method was validated on benchmark functions and applied to multiclass segmentation of COVID-19 X-ray images from the Kaggle Radiography database.

Main Results:

• The proposed chance-based birds' intelligence optimizer demonstrated superior performance compared to seagull/cuckoo optimization.
• Statistical analysis using Friedman's mean rank test ranked the proposed method first among compared techniques.
• Significant improvements were observed in Peak Signal to Noise Ratio (PSNR), Feature Similarity Index (FSIM), and Structure Similarity Index (SSIM), with an average PSNR increase of approximately 11%.

Conclusions:

• The novel non-entropy-based thresholding method offers a significant improvement for segmenting COVID-19 X-ray images.
• The proposed chance-based birds' intelligence optimizer is effective for image segmentation tasks.
• This method holds promise for advancing medical image analysis and COVID-19 diagnosis.