Single Detect Focused YOLO Framework for Robust Mitotic Figure Detection

Creative Commons License

TOPUZ Y., GÖKCAN M. T., Yıldız S., VARLI S.

3rd MICCAI challenge on Mitosis Domain Generalization, MIDOG 2025, addressing WSI-level mitotic figure detection and atypia classification, held in conjunction with 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025, Daejeon, Güney Kore, 23 Eylül 2025, cilt.16519 LNCS, ss.65-74, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 16519 LNCS
  • Doi Numarası: 10.1007/978-3-032-25180-0_7
  • Basıldığı Şehir: Daejeon
  • Basıldığı Ülke: Güney Kore
  • Sayfa Sayıları: ss.65-74
  • Anahtar Kelimeler: Domain shift, MIDOG, Small object detection
  • Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

Mitotic figure detection is a crucial task in computational pathology, as mitotic activity serves as a strong prognostic marker for tumor aggressiveness. However, domain variability that arises from differences in scanners, tissue types, and staining protocols poses a major challenge to the robustness of automated detection methods. In this study, we introduce SDF-YOLO (Single Detect Focused YOLO), a lightweight yet domain-robust detection framework specifically designed for small and rare targets such as mitotic figures. Our model incorporates a single detection head aligned with the mitotic figure scale, coordinate attention to enhance positional sensitivity, and improved cross-channel feature mixing. We conducted experiments on three datasets that span human and canine tumors: MIDOG++, canine cutaneous mast cell tumor (CCMCT), and canine mammary carcinoma (CMC). In the preliminary test phase of the MIDOG2025 challenge, we achieved an F1 score of 0.766, demonstrating the competitive accuracy and computational efficiency of our approach. In the final test phase, our model obtained an F1 score of 0.709, placing us 3rd overall in the competition. Notably, we achieved the highest FROC-AUC score (5.347) among all participants, underscoring the robustness and strong domain generalization capabilities of SDF-YOLO. These results indicate that the SDF-YOLO framework provides a reliable and efficient solution for robust mitotic figure detection across diverse domains.