Detection of Necrosis and Steatosis in Liver Histopathology Images Using a CARAFE- and Wise-IoU-Enhanced YOLOv8
International Journal of Advanced Computer Science and Applications · 2026 · 17 (10)
DOI: 10.14569/ijacsa.2026.0171058 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Advanced Computer Science and Applications, 2026. Verfasst von Lailil Muflikhah und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Automated localization of necrosis and steatosis in liver histopathology images is challenging because tissue regions exhibit heterogeneous appearances and indistinct boundaries. This preliminary study evaluates the integration of two established components, Content-Aware ReAssembly of FEatures (CARAFE) and Wise-IoU v3, into YOLOv8s for detecting necrosis, steatosis, and normal cellular regions. CARAFE replaces nearest-neighbor upsampling in the neck to preserve fine-grained morphological details, while Wise-IoU v3 replaces the CIoU bounding-box regression loss to attenuate harmful gradients from ambiguous lesion boundaries. The model was initialized using PanNuke-pretrained weights and evaluated on 49 annotated liver biopsy images (6,126 total bounding boxes) partitioned at the slide level into a 70:15:15 training, validation, and testing split (34, 7, and 8 images; 1,007 test instances) across five independent random seeds. Across the five runs, the modified model achieved 90.18 ± 0.24% precision, 89.56 ± 0.28% recall, 89.87 ± 0.25% F1-score, and 91.08 ± 0.22% mAP@50 (95% CI: 90.89%–91.27%) on the held-out test set, outperforming baseline YOLOv8s (90.18 ± 0.29% mAP@50) with only 0.3 million additional parameters. Component ablation further confirmed the complementary benefits of PanNuke domain initialization, CARAFE upsampling, and Wise-IoU v3 optimization over standard COCO weights. Although multi-seed evaluation over 1,007 test instances confirms consistent run-to-run gains, the small single-source cohort remains a preliminary constraint requiring future multi-center external validation.
Zitieren
APA
Muflikhah, L., & Wardhana, P. W. K. (2026). Detection of Necrosis and Steatosis in Liver Histopathology Images Using a CARAFE- and Wise-IoU-Enhanced YOLOv8. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/ijacsa.2026.0171058
Vancouver
Muflikhah L, Wardhana PWK. Detection of Necrosis and Steatosis in Liver Histopathology Images Using a CARAFE- and Wise-IoU-Enhanced YOLOv8. International Journal of Advanced Computer Science and Applications. 2026;17(10). doi:10.14569/ijacsa.2026.0171058
BibTeX
@article{Muflikhah2026,
author = {Muflikhah, Lailil and Wardhana, Putu Wahyu Kusuma},
title = {Detection of Necrosis and Steatosis in Liver Histopathology Images Using a CARAFE- and Wise-IoU-Enhanced YOLOv8},
journal = {International Journal of Advanced Computer Science and Applications},
year = {2026},
volume = {17},
number = {10},
doi = {10.14569/ijacsa.2026.0171058}
}
EUVIMED