To segment or not to segment: COVID-19 detection for chest X-rays.

Area of Science:

• Artificial intelligence in medical imaging
• Computer vision for disease detection
• Machine learning and deep learning applications

Background:

• Artificial intelligence (AI) is increasingly integrated into technology, with medical imaging being a key application area.
• Computer vision (CV) algorithms show promise in analyzing medical images and recognizing patterns.
• AI has become integral to state-of-the-art medical imaging, improving outcomes.

Purpose of the Study:

• To investigate computer vision (CV) segmentation techniques for COVID-19 analysis using chest X-rays (CXRs).
• To compare the effectiveness of k-means, U-net, and flood fill segmentation methods for lung region extraction.
• To evaluate the performance of machine learning (ML) and deep learning (DL) models for identifying COVID-19 lesions.

Main Methods:

• Utilized k-means, U-net, and flood fill for lung region segmentation in CXRs.
• Employed ML and DL models to detect COVID-19 lesion molecules in healthy and pathological lung images.
• Evaluated ML and DL performance in conjunction with CV segmentation techniques.
• Tested DL model robustness against real-world noise, including salt and pepper noise.

Main Results:

• CV segmentation techniques showed lower performance compared to direct ML and DL models.
• Optimal AI algorithms achieved accuracy between 0.92-0.94.
• Incorporating CV algorithms reduced accuracy to approximately 0.81-0.88.
• Real-world noise negatively impacted DL model performance.

Conclusions:

• Direct application of ML and DL models is more effective for COVID-19 lesion detection than using CV segmentation preprocessing.
• CV segmentation techniques do not enhance, and may hinder, the performance of AI models in this context.
• Further research is needed to optimize AI models for robust performance under noisy conditions in medical imaging.