Dual-domain radio-frequency signal and image joint modeling for coronary calcium detection in intravascular ultrasound.

Coronary artery disease remains a leading cause of mortality worldwide. Accurate detection and angular quantification of coronary calcification are important for optimizing percutaneous coronary intervention. Intravascular Ultrasound provides high-resolution visualization of plaques, where calcification typically appears as hyperechoic regions with acoustic shadowing. However, grayscale intravascular ultrasound is sensitive to imaging conditions and reconstruction algorithms, making automated calcification analysis challenging. This study aims to automate the detection of calcification and the quantification of angles. We propose a dual-branch deep learning framework that integrates raw intravascular ultrasound radio-frequency (RF) signals with polar-coordinate grayscale images. A physics-consistent complex convolution module is introduced to preserve the intrinsic physical structure of RF signals during feature extraction. The framework further incorporates phase-amplitude guided dual-path cross-attention and a shadow-aware radial A-line classifier to enhance RF-image interaction and angle estimation. Calcification angles are computed from the predicted segmentation combined with lumen centroid localization. Experiments were conducted on a multicenter development dataset from four centers, including 78 patients, 136 pullbacks, and 32,672 IVUS frames. An independent external clinical test set comprising 10 patients, 15 pullbacks, and 2,362 frames was further used for final evaluation. On the internal validation set, the proposed method achieved a Dice coefficient of 0.921 and a Hausdorff distance of 0.251 mm for calcification segmentation, with an A-line classification accuracy of 97.1% for calcification angle quantification. The model also maintained stable performance on the external clinical test set, supporting its potential generalizability in real-world clinical scenarios. Joint modeling of RF signals and grayscale intravascular ultrasound features enables accurate and robust calcification detection and angular quantification, supporting objective and reproducible clinical assessment.