Radioport: a radiomics-reporting network for interpretable deep learning in BI-RADS classification of mammographic...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Informatics

Background:

• Deep learning radiomics (DLR) is often a "black-box" for radiologists, lacking explainability.
• Automatic generation of diagnostic reports can improve the semantic explanation of DLR.
• Mammographic calcification diagnosis requires clear and interpretable findings.

Purpose of the Study:

• To develop an interpretable deep learning model for mammographic calcification diagnosis.
• To enhance the explainability of deep learning radiomics (DLR) through report generation.
• To improve the clarity and readability of generated medical reports.

Main Methods:

• Proposed a novel radiomics-reporting network (Radioport) incorporating text attention.
• Utilized convolutional neural networks to extract radiomic features from mammograms.
• Mapped visual radiomic features to textual features for diagnostic report generation.

Main Results:

• Demonstrated semantic enhancement of DLR interpretability.
• Showcased improved readability of generated medical reports.
• Validated effectiveness on a breast calcification dataset.

Conclusions:

• The developed interpretable textual model can simulate the mammographic calcification diagnosis process.
• Radioport offers a potential solution to the explainability challenge in DLR.
• The model aids radiologists by providing clearer, semantically enhanced diagnostic reports.