Lumen Estimation using YOLOv8 with Falling-Ball Algorithms in Polar-Space

Accurate lumen contour detection in Intravascular Ultrasound (IVUS) images is critical for diagnosing cardiovascular diseases and guiding interventions. However, traditional methods often struggle with the noise, low contrast, and artifacts present in IVUS images. This paper proposes a novel method for lumen estimation using a Polar-Space Falling-Ball Algorithm, combined with deep learning and computational geometry techniques. The method begins with lumen and media-adventitia border detection using the YOLOv8 deep learning model, which is fine-tuned for IVUS images. Power-Law Gamma Transformation is applied to enhance image contrast, and the IVUS image is transformed into polar coordinates for easier border analysis. The Falling-Ball Algorithm is employed to refine the detected borders, followed by high-order polynomial regression to estimate the lumen contour. The final contours are then converted back to Cartesian coordinates for visualization. Experimental results on a dataset of 203 IVUS images show that the proposed method achieves a significantly improving accuracy over traditional approaches. This method offers a robust, accurate, and computationally efficient solution for IVUS lumen estimation, with the potential for real-time clinical application.