We developed a fully automated, two-step deep learning approach for characterizing coronary calcified plaque in intravascular optical coherence tomography (IVOCT) images. First, major calcification lesions were detected from an entire pullback using a 3D convolutional neural network (CNN). Second, a SegNet deep learning model with the Tversky loss function was used to segment calcified plaques in the major calcification lesions. The fully connected conditional random field and the frame interpolation of the missing calcification frames were used to reduce classification errors. We trained/tested the networks on a large dataset comprising 8,231 clinical images from 68 patients with 68 vessels and 4,320 ex vivo cadaveric images from 4 hearts with 4 vessels. The 3D CNN model detected major calcifications with high sensitivity (97.7%), specificity (87.7%), and F1 score (0.922). Compared to the standard one-step approach, our two-step deep learning approach significantly improved sensitivity (from 77.5% to 86.2%), precision (from 73.5% to 75.8%), and F1 score (from 0.749 to 0.781). We investigated segmentation performance for varying numbers of training samples; at least 3,900 images were required to obtain stable segmentation results. We also found very small differences in calcification attributes (e.g., angle, thickness, and depth) and identical calcium scores on repetitive pullbacks, indicating excellent reproducibility. Applied to new clinical pullbacks, our method has implications for real-time treatment planning and imaging research.

Major calcifications are of great concern when performing percutaneous coronary intervention (PCI) because they can hinder stent deployment. Approximately 700,000 PCIs are performed each year, and many involve the use of stents to open up obstructed coronary arteries.¹ Calcified plaques are found in 17% to 35% of patients undergoing PCI.^{2 – 4} Calcifications can lead to stent underexpansion and strut malapposition, which in turn can lead to increased risk of thrombosis and in-stent restenosis.^{5 – 10} A cardiologist has several options when confronting a calcified lesion: high balloon pressures (up to 30 atm) to fracture the calcification, scoring balloon, Shockwave^TM intravascular lithotripsy (IVL), rotational atherectomy, etc. In some cases, the lesion may not be treatable.

Deep learning based methods are routinely used to segment various structures of interest in varied medical imaging modalities. Acquiring annotations for a large number of images requires a skilled analyst, and the process is both time consuming and challenging. Our approach to reduce effort is to reduce the number of images needing detailed annotation. For intravascular optical coherence tomography (IVOCT) image pullbacks, we tested 10% to 100% of training images derived from two schemes: equally-spaced image subsampling and deep-learning- based image clustering. The first strategy involves selecting images at equally spaced intervals from the volume, accounting for the high spatial correlation between neighboring images. In clustering, we used an autoencoder to create a deep feature space representation, performed k-medoids clustering, and then used the cluster medians for training. For coronary calcifications, a baseline U-net model was trained on all images from volumes of interest (VOIs) and compared with fewer images from the sub-sampling strategies. For a given sampling ratio, the clustering based method performed better or similar as compared to the equally spaced sampling approach (e.g., with 10% of images, mean F1 score for calcific class increased from 0.52 to 0.63, with equal spacing and with clustering, respectively). Additionally, for a fixed number of training images, sampling images from more VOIs performed better than otherwise. In conclusion, we recommend the clustering based approach to annotate a small fraction of images, creating a baseline model, which potentially can be improved further by annotating images selected using methods described in active learning research.

It can be difficult/impossible to fully expand a coronary artery stent in a heavily calcified coronary artery lesion. Under-expanded stents are linked to later complications. Here we used machine/deep learning to analyze calcifications in pre-stent intravascular optical coherence tomography (IVOCT) images and predicted the success of vessel expansion. Pre- and post-stent IVOCT image data were obtained from 110 coronary lesions. Lumen and calcifications in pre-stent images were segmented using deep learning, and lesion features were extracted. We analyzed stent expansion along the lesion, enabling frame, segmental, and whole-lesion analyses. We trained regression models to predict the post-stent lumen area and then computed the stent expansion index (SEI). Best performance (root-mean-square-error = 0.04 ± 0.02 mm²,r = 0.94 ± 0.04,p < 0.0001) was achieved when we used features from both lumen and calcification to train a Gaussian regression model for segmental analysis of 31 frames in length. Stents with minimum SEI > 80% were classified as “well-expanded;” others were “under-expanded.” Under-expansion classification results (e.g., AUC = 0.85 ± 0.02) were significantly improved over a previous, simple calculation, as well as other machine learning solutions. Promising results suggest that such methods can identify lesions at risk of under-expansion that would be candidates for intervention lesion preparation (e.g., atherectomy).

Microvessels in vascular plaque are associated with plaque progression and are found in plaque rupture and intra-plaque hemorrhage. To analyze this characteristic of vulnerability, we developed an automated deep learning method for detecting microvessels in intravascular optical coherence tomography (IVOCT) images. A total of 8403 IVOCT image frames from 85 lesions and 37 normal segments were analyzed. Manual annotation was performed using a dedicated software (OCTOPUS) previously developed by our group. Data augmentation in the polar (r,θ) domain was applied to raw IVOCT images to ensure that microvessels appear at all possible angles. Pre-processing methods included guidewire/shadow detection, lumen segmentation, pixel shifting, and noise reduction. DeepLab v3+ was used to segment microvessel candidates. A bounding box on each candidate was classified as either microvessel or non-microvessel using a shallow convolutional neural network. For better classification, we used data augmentation (i.e., angle rotation) on bounding boxes with a microvessel during network training. Data augmentation and pre-processing steps improved microvessel segmentation performance significantly, yielding a method with Dice of 0.71 ± 0.10 and pixel-wise sensitivity/specificity of 87.7 ± 6.6%/99.8 ± 0.1%. The network for classifying microvessels from candidates performed exceptionally well, with sensitivity of 99.5 ± 0.3%, specificity of 98.8 ± 1.0%, and accuracy of 99.1 ± 0.5%. The classification step eliminated the majority of residual false positives and the Dice coefficient increased from 0.71 to 0.73. In addition, our method produced 698 image frames with microvessels present, compared with 730 from manual analysis, representing a 4.4% difference. When compared with the manual method, the automated method improved microvessel continuity, implying improved segmentation performance. The method will be useful for research purposes as well as potential future treatment planning.