An Attention Mechanism with Multiple Knowledge Sources for COVID-19 Detection from CT Images

The system uses attention mechanisms to focus on relevant areas: The researchers integrated infected regions and heat maps with full CT images through an attention mechanism, which helps the network concentrate on local lesion areas rather than processing the entire image equally—similar to how doctors visually focus on suspicious regions when reading scans.

Multiple information sources improve diagnostic robustness: By combining several types of useful information relevant to clinical judgment, the approach demonstrated superior performance compared to baseline methods and made the system more resistant to noise in the images, which is important for real-world medical applications.

Explainability is built into the model design: The researchers specifically highlight that their "grey-box model" provides explainable features, allowing doctors to understand the connection between what the network sees in the input image and its diagnostic output—addressing a key concern in responsible AI about medical decision-making transparency.

The approach mimics clinical reasoning: Rather than simply introducing novel architectures or collecting larger datasets, the strategy deliberately incorporates how doctors actually make judgments when interpreting CT scans, suggesting that AI systems designed with clinical workflows in mind may be more trustworthy and effective.