| Nooshin Nemati, Emrah Hancer, Akın Atakan Goren, Refik Samet A Comparative Study of YOLO-based Architectures for Mitosis Detection in Histopathology Images |
|---|
| Abstract. Mitosis detection is a crucial task in tumor grading and prognosis assessment, especially in breast cancer diagnosis. However, manual identification of mitotic cells in Hematoxylin and Eosin (H&E) stained histopathological images is time-consuming, subjective, and labor-intensive. Automated computer-aided diagnosis systems can significantly support pathologists by improving efficiency and consistency. In this study, we present a comprehensive comparative evaluation of several YOLO-based one-stage deep learning architectures for mitosis detection. Specifically, YOLOv8, YOLOX, YOLOv11, and YOLOv26 are assessed on three publicly available datasets: ICPR12, ICPR14, and MiDeSeC. The models are evaluated using precision, recall, and F1-score to ensure a balanced performance comparison. Experimental results indicate that YOLOv8 consistently achieves strong performance across all datasets, demonstrating robustness and reliability. Overall, the results suggest that modern YOLO-based detectors provide competitive performance for automated mitosis detection and represent a promising direction for further research and practical development. |
| Keywords: Mitosis detection, Histopathological image, Object detection, YOLO |
 Download PDF |
| DOI: https://doi.org/10.54381/itta2026.2.09 |