Angiography-Based Detection of Coronary Artery Stenosis Using YOLO Algorithm


Creative Commons License

Küpeli H., Kuru İ., Çevik K. K., Bozkurt A.

TRAITEMENT DU SIGNAL, cilt.42, sa.3, ss.1525-1539, 2025 (SCI-Expanded)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 42 Sayı: 3
  • Basım Tarihi: 2025
  • Doi Numarası: 10.18280/ts.420325
  • Dergi Adı: TRAITEMENT DU SIGNAL
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, PASCAL, Business Source Elite, Business Source Premier, Compendex, zbMATH
  • Sayfa Sayıları: ss.1525-1539
  • Akdeniz Üniversitesi Adresli: Evet

Özet

This study investigates the efficacy of You Only Look Once (YOLO) algorithms in detecting coronary artery stenosis from angiographic images. The dataset utilized comprises 8,325 grayscale images sourced from publicly available databases, featuring patients diagnosed with single-vessel coronary artery disease. An expert cardiologist annotated the images to precisely mark areas of vascular occlusion, providing reliable training data. Four distinct datasets were constructed and divided into training (80%) and testing (20%) subsets. YOLO v5, v7, and v8 models were trained over 100 epochs to evaluate their performance in identifying stenotic regions. The study emphasizes the advantages of YOLO algorithms, particularly their ability to detect multiple objects in real-time with high accuracy, due to their single-stage detection architecture. Performance metrics such as Mean Average Precision (MAP), precision, recall, and F1-score were computed to assess model effectiveness. The results demonstrate that YOLO v5 and YOLO v8 provide robust detection capabilities, outperforming YOLO v7, especially in complex image scenarios. This research highlights the potential integration of YOLO models in clinical workflows, offering a rapid and accurate tool for automated analysis of coronary artery stenosis.

This study investigates the efficacy of You Only Look Once (YOLO) algorithms in detecting coronary artery stenosis from angiographic images. The dataset utilized comprises 8,325 grayscale images sourced from publicly available databases, featuring patients diagnosed with single-vessel coronary artery disease. An expert cardiologist annotated the images to precisely mark areas of vascular occlusion, providing reliable training data. Four distinct datasets were constructed and divided into training (80%) and testing (20%) subsets. YOLO v5, v7, and v8 models were trained over 100 epochs to evaluate their performance in identifying stenotic regions. The study emphasizes the advantages of YOLO algorithms, particularly their ability to detect multiple objects in real-time with high accuracy, due to their single-stage detection architecture. Performance metrics such as Mean Average Precision (MAP), precision, recall, and F1-score were computed to assess model effectiveness. The results demonstrate that YOLO v5 and YOLO v8 provide robust detection capabilities, outperforming YOLO v7, especially in complex image scenarios. This research highlights the potential integration of YOLO models in clinical workflows, offering a rapid and accurate tool for automated analysis of coronary artery stenosis.