Comparative Analysis of YOLOv11 and YOLOv9 for Eye Regions Segmentation G z B lgeleri Segmentasyonu i in YOLOv11 ve YOLOv9'un Karsilastirmali Analizi

Nusari A. N., POLAT M., ORAL E. A., ÖZBEK İ. Y.

33rd IEEE Conference on Signal Processing and Communications Applications, SIU 2025, İstanbul, Türkiye, 25 - 28 Haziran 2025, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu66497.2025.11112172
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: Computer Vision, deep learning, Eye Segmentation, Segmentation, YOLOv11, YOLOv9
  • Atatürk Üniversitesi Adresli: Evet

Özet

accurate segmentation of the retina, optic discs, and other anatomical structures of the eye from eye images is crucial for ophthalmic diagnosis and the evaluation of eye diseases. In previous studies, the DeepLabv3 model was used for segmenting eye regions, achieving an average Dice score of 81%. In this study, the latest YOLO models, YOLOv9 and YOLOv11, were utilized due to their potential to achieve an optimal balance between accuracy and efficiency in medical image analysis.Experimental results demonstrate that YOLOv11n outperforms other models. YOLOv11n achieved MAP50, MAP50-95, and Dice scores of 90.6%, 49.3%, and 90%, respectively. In contrast, YOLOv9-C obtained MAP50, MAP50-95, and Dice scores of 89%, 47.6%, and 88.9%, respectively.These findings indicate that the YOLOv11n model improves segmentation accuracy for eye images, demonstrating its effectiveness compared to previous models.