A deep learning methodology for fully-automated quantification of calcific burden in high-resolution intravascular ultrasound images

Comprehensive Summary

He and their colleagues evaluated whether a deep learning method could accurately detect and quantify coronary calcific burden from high resolution intravascular ultrasound images. The study analyzed IVUS data from 78 patients, including 197 vessels for model training and 30 vessels for testing. In total, 26,211 frames were used for training and 5,138 for validation. The deep learning model demonstrated strong agreement with expert analysts for the presence of calcium, with high correlation for calcium arc and a calcific burden index at both lesion and segment levels (r ≥ 0.97). Performance for identifying calcific rich lesions with calcium arc ≥90 degrees was high, with sensitivity 95.5 percent and an overall accuracy 91.7 percent. This automated analysis reduced processing time from approximately 1.5 hours per vessel to under one minute.

Outcomes and Implications

This study suggests that deep learning based IVUS analysis can provide fast and reproducible quantification of coronary calcification with performance comparable to expert readers. In practice, integrating this approach into existing IVUS workflows is very likely to improve PCI planning by identifying calcific burden and distribution without increasing analysis time or requiring specialized expertise. With such a decrease in analysis time, there is immediate opportunity for implementation. Further validation across centers and IVUS systems could improve model generalizability before routine clinical adoption.