Facilitating COVID recognition from X-rays with computer vision models and transfer learning.

Area of Science:

• Medical imaging analysis
• Computer vision in healthcare
• Machine learning applications

Background:

• Electronic Health Records (EHR) increasingly utilize multimedia data, including complex medical images and videos.
• Accurate and efficient recognition of COVID-19 is crucial for patient management and public health.
• Existing computer vision models require substantial data for training, posing challenges for rare or emerging diseases.

Purpose of the Study:

• To investigate the efficacy of transfer learning with computer vision for COVID-19 detection using chest X-rays.
• To determine the optimal computer vision models and data augmentation strategies for accurate COVID-19 recognition from limited image datasets.
• To achieve maximum diagnostic accuracy with a minimum number of training and validation samples.

Main Methods:

• Transfer learning was applied to a large dataset of publicly available chest X-rays.
• Computer vision models were adapted using data augmentation techniques.
• Model performance was evaluated by adjusting training and validation sample sizes to find optimal parameters for accuracy and efficiency.

Main Results:

• Transfer learning effectively utilizes limited chest X-ray data for COVID-19 identification.
• Data augmentation strategies improved model adaptability and performance.
• The study identified optimal models and sample sizes for achieving high accuracy in COVID-19 detection.

Conclusions:

• Combining chest radiography with transfer learning offers a promising, cost-effective method for improving the accuracy and timeliness of COVID-19 radiological interpretation.
• This approach has significant potential for application during COVID-19 diagnosis and recovery phases.
• Further research can explore advanced multimedia analysis and machine learning techniques in healthcare.