Intracoronary Optical Coherence Tomography Image Processing and Vessel Classification Using Machine Learning

Intracoronary Optical Coherence Tomography (OCT) enables high-resolution visualization of coronary vessel anatomy but presents challenges due to noise, imaging artifacts, and complex tissue structures. This paper proposes a fully automated pipeline for vessel segmentation and classification in OCT images using machine learning techniques. The proposed method integrates image preprocessing, guidewire artifact removal, polar-to-Cartesian transformation, unsupervised K-means clustering, and local feature extraction. These features are used to train Logistic Regression and Support Vector Machine classifiers for pixel-wise vessel classification. Experimental results demonstrate excellent performance, achieving precision, recall, and F1-score values up to 1.00 and overall classification accuracy of 99.68%. The proposed approach provides accurate vessel boundary detection while maintaining low computational complexity and requiring minimal manual annotation. This method offers a reliable and efficient solution for automated OCT image analysis and has potential applications in clinical decision support and real-time medical image processing.

💡 Research Summary

The paper presents a fully automated pipeline for processing intracoronary optical coherence tomography (OCT) images and classifying each pixel as vessel or background using conventional machine‑learning techniques. The authors begin by addressing the inherent challenges of OCT data: speckle noise, salt‑and‑pepper impulse noise, and the guidewire shadow that frequently obscures a portion of the lumen. Noise level is quantified through the standard deviation of the Laplacian operator and a salt‑and‑pepper ratio; images with high noise are filtered with a 3×3 median filter to preserve edges while suppressing outliers. Guidewire artifacts are removed by locating the column of minimum intensity, shifting the vessel wall back into this region, and applying a blend‑and‑crop operation to create a seamless transition.

Because raw OCT frames are acquired in polar coordinates, the pipeline converts them to Cartesian space, straightening the circular vessel cross‑section and facilitating subsequent processing steps. In the transformed domain, an unsupervised K‑means clustering (two clusters) separates the image into candidate vessel and background regions without any manual annotation. The clustering is repeated with multiple random initializations to mitigate sensitivity to seed selection.

From each pixel, a local 11×11 sliding window is extracted, and nine statistical and texture descriptors (mean, variance, entropy, contrast, energy, homogeneity, correlation, etc.) are computed. These features capture subtle intensity variations and textural patterns that differentiate lumen from surrounding tissue. The feature vectors are then fed into two supervised classifiers: a logistic regression model with a linear decision boundary and a support vector machine (SVM) employing a radial‑basis‑function kernel. Both models are trained on a modest dataset of several hundred OCT frames, using five‑fold cross‑validation to assess performance.

Results show exceptionally high metrics: precision, recall, and F1‑score reach 1.00 (or very close) and overall classification accuracy attains 99.68 %. The SVM slightly outperforms logistic regression at vessel boundaries, reducing false positives in regions with ambiguous texture. Computationally, the entire pipeline processes a single frame in under one second on a standard CPU, indicating feasibility for real‑time clinical decision support.

The authors acknowledge limitations, notably the small, predominantly normal‑vessel dataset and the absence of pathological cases such as plaques or calcifications. Consequently, the reported performance may not fully generalize to more diverse clinical scenarios. Future work is outlined to include larger, multi‑center datasets, integration of deep‑learning based feature extraction (e.g., convolutional autoencoders), and real‑time deployment within catheter‑lab workflows. By combining simple, interpretable preprocessing with lightweight machine‑learning classifiers, the study offers a low‑complexity yet highly accurate solution for automated OCT vessel segmentation, paving the way for broader adoption in interventional cardiology.