COVID-19 Detection Through Transfer Learning Using Multimodal Imaging Data.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Disease Detection
• Radiology and Diagnostic Imaging

Background:

• Early detection of COVID-19 is crucial for treatment and containment.
• Medical imaging modalities like X-ray, Ultrasound, and CT scans are vital for diagnosis.
• Developing accurate AI models for COVID-19 detection is challenging due to limited, variable datasets.

Purpose of the Study:

• To investigate the efficacy of transfer learning with deep learning models for COVID-19 detection across X-ray, Ultrasound, and CT scans.
• To provide medical professionals with an AI-powered 'second opinion' for faster diagnosis.
• To address challenges associated with small and low-quality COVID-19 datasets.

Main Methods:

• Comparative study of various Convolutional Neural Network (CNN) models to identify optimal architecture.
• Optimization of the VGG19 model using transfer learning for COVID-19 detection on diverse medical images.
• Implementation of an image pre-processing stage to enhance dataset quality and reduce noise.

Main Results:

• Ultrasound imaging demonstrated superior COVID-19 detection accuracy (100%) compared to X-ray (86%) and CT scans (84%).
• The optimized VGG19 model achieved considerable detection performance across all three imaging modalities.
• Deep learning models struggled with limited data, highlighting the importance of data quality and pre-processing.

Conclusions:

• Transfer learning with optimized CNN models, particularly VGG19, offers a viable approach for COVID-19 detection using medical imaging.
• Ultrasound emerges as a highly accurate modality for AI-assisted COVID-19 diagnosis.
• Data pre-processing and careful model selection are critical for reliable deep learning applications in medical diagnostics with scarce data.