BackgroundThe presence of extensive calcification is a primary concern when planning and implementing a vascular percutaneous intervention such as stenting. If the balloon does not expand, the interventionalist must blindly apply high balloon pressure, use an atherectomy device, or abort the procedure. Intravascular optical coherence tomography can evaluate extensive calcification and aid intervention planning, there we aim to develop a method for automatic detection of calcium in coronary IVOCT images.

A patch-wise classification deep learning model was developed and trained. IVOCT pullbacks from 2 medical centers were analyzed in a core lab, annotating basic calcified components and other structures. We divided our dataset into training, validation, and test dataset. We used polar transform to split images with different angle to patches (30, 45, 60 and 90 degrees).Figure 1 showed we used polar transform to split our images and  we demonstrated a splited image with 45 degrees in figure 2.

Annotated IVOCT images from 111 pullbacks (111 patients) were divided into 1871 frames of normal images and 3951 frames of calcified lesions (total frames: 5822 frames).We used polar transform to split images with different angle to patches. We found the images with 90 degrees splitting have better performance. The sensitivity was 0.901, the specificity was 0.8847 and the accuracy was 0.892.           

A novel artificial intelligence framework for automatic calcified lesion detection in IVOCT was developed, providing excellent diagnostic accuracy. This model might reduce subjectivity in image interpretation with potential applications in research and IVOCT-guided percutaneous coronary intervention with calcified lesions.