Multi-modal trained artificial intelligence solution to triage chest X-ray for COVID-19 using pristine ground-truth,...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed tomography (CT) and X-ray are crucial for COVID-19 patient triage.
• Thoracic CT offers higher sensitivity than X-ray for COVID-19 diagnosis.
• CT accessibility limitations necessitate alternative solutions like AI-assisted X-ray analysis.

Purpose of the Study:

• To develop and evaluate an AI model for COVID-19 triage using single chest X-ray images.
• To investigate the benefits of multi-modal training (X-ray and CT data) for AI model performance.
• To compare the AI model's diagnostic and segmentation capabilities against experienced radiologists.

Main Methods:

• Developed a functional AI model for classifying and segmenting COVID-19 from chest X-rays.
• Trained the AI model using a multi-modal approach incorporating both X-ray and CT data.
• Conducted a reader study to compare AI performance against radiologists and evaluated generalization on independent datasets.

Main Results:

• Multi-modal training improved binary classification AUC from 0.89 to 0.93 and Dice coefficient for pathology localization from 0.59 to 0.62.
• AI model achieved comparable or superior performance to radiologists in segmentation (Dice 0.52-0.55 vs. 0.53) and classification (AUC 0.93 vs. 0.87/0.81).
• The proposed method demonstrated superior performance compared to state-of-the-art methods in generalization studies.

Conclusions:

• Multi-modal training significantly enhances AI model performance for COVID-19 detection and localization using X-ray images.
• The AI model shows potential as a reliable tool for widespread clinical use in COVID-19 triage, comparable to expert radiologists.
• Leveraging multi-modal data during training benefits single-modal inferencing, offering a promising direction for AI in medical diagnostics.