TL;DR
This paper introduces two novel methods for detecting coronary artery stenosis from angiograms: an automatic method for global artery tree analysis and an interactive method for detailed segment analysis. These methods aim to improve the accuracy and clinical utility of CAD diagnosis by addressing limitations in existing approaches, particularly the lack of precise, segment-specific stenosis evaluation.
What Problem Does This Paper Address?
Coronary artery disease (CAD) is a major global health concern, and coronary angiography (CA) is the gold standard for its diagnosis. However, current methods for identifying and quantifying coronary artery stenosis are largely subjective, leading to inefficiencies and potential errors. Existing automated methods often focus on global analysis and lack the ability to perform detailed, quantitative assessments of specific vessel segments, which is crucial for clinical practice. Furthermore, there's a gap in research regarding the detection of stenoses within prespecified vessel segments. This paper aims to bridge this gap by proposing a two-pronged approach that combines automatic and interactive stenosis detection.
Methods
The paper proposes two main methods:
- Automatic Stenosis Detection: This method first preprocesses the angiogram images using a combination of multiscale image enhancement, denoising (ROF), sharpness enhancement (UM), and contrast enhancement (CLAHE). It then extracts the coronary artery tree using an Improved Adaptive Geometrical Tracking Algorithm (IAGVT). This algorithm builds upon existing tracking methods by incorporating a C-V model for more robust vessel contour extraction and adjustment. After segmenting the vessels, it measures diameters and evaluates stenotic degrees using a discriminant function. A key innovation here is the inclusion of a bifurcation detection process to ensure the entire vessel tree is captured, overcoming limitations of simple tracking.
- Interactive Stenosis Detection: This method allows clinicians to select a specific vessel segment of interest by specifying start and end points. To extract this segment, a novel potential energy function is introduced. This function guides the tracking process, preventing deviations at bifurcations and ensuring the path approaches the specified endpoint. Once the segment is extracted, the same diameter measurement and stenosis evaluation algorithms from the automatic method are applied for precise, quantitative analysis.
Datasets
The proposed methods were evaluated on a dataset consisting of 53 grayscale coronary angiogram images of size 400x400 pixels. These images were sourced from DICOM files of 30 patient cases provided by Qilu Hospital (Qingdao). The ground truth for stenosis detection was provided by an expert. No new benchmark datasets were introduced.
Key Insights and Ideas
- The Need for Interactive Analysis: The paper strongly emphasizes that while automatic detection is useful for screening, clinical practice often demands detailed, segment-specific analysis. The interactive method directly addresses this by empowering clinicians to focus on areas of concern.
- Combining Global and Local Approaches: The proposed two-stage approach (automatic for screening, interactive for detailed analysis) offers a powerful synergy, leveraging the strengths of both.
- Improved Vessel Tracking: The IAGVT, with its incorporation of the C-V model and robust bifurcation detection, represents a significant improvement over existing tracking algorithms, leading to more accurate vessel centerline extraction.
- Potential Energy for Focused Tracking: The introduction of a potential energy function for interactive segment extraction is a clever way to guide the tracking algorithm towards a desired segment, especially in complex vascular structures.
Key Results
- Automatic Stenosis Detection Performance: The automatic method achieved a precision of 0.821, sensitivity of 0.757, and an F₁ score of 0.788.
- Comparison with Existing Methods: The proposed automatic method demonstrated superior performance compared to Xiao's method [8] in terms of sensitivity, precision, and F₁-score.
- Interactive Method's Precision: The interactive method was shown to provide more precise outcomes, particularly for stenoses missed by the automatic method.
- Diameter Measurement Accuracy: The proposed diameter measurement method showed significantly lower relative errors compared to Xiao's method, with visual results indicating closer alignment with expert-marked ground truth.
Limitations and Future Work
- Preprocessing Dependence: The authors acknowledge that the methods have a high dependence on the preprocessing and vessel contour extraction algorithms, which can lead to errors in complex vessel structures.
- Future Work: The authors plan to explore more advanced methods, such as deep learning, for preprocessing and vessel contour extraction to further enhance detection accuracy. They also aim to improve the robustness of their algorithms for more complex vascular structures.
Summary updated