Quantitative Evaluation of Nonylphenol Nephrotoxicity from Histopathological Images: A Multidimensional Radiomics and Machine Learning Framework


Şentürk T., Latifoğlu F., Yalçın B., Göç R., Göktepe Ö., Önder G. Ö., ...Daha Fazla

APPLIED SCIENCES, cilt.16, sa.22, ss.1-31, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 16 Sayı: 22
  • Basım Tarihi: 2026
  • Doi Numarası: 10.3390/app162211326
  • Dergi Adı: APPLIED SCIENCES
  • Derginin Tarandığı İndeksler: Applied Science & Technology Source, Scopus, Science Citation Index Expanded (SCI-EXPANDED), Compendex, INSPEC, Directory of Open Access Journals
  • Sayfa Sayıları: ss.1-31
  • Erciyes Üniversitesi Adresli: Evet

Özet

Background/Objectives: Nonylphenol (NP), an endocrine disruptor, damages the kidney, but histological assessment remains largely visual. We tested whether image texture separates NP dose groups under block-grouped validation. Methods: Kidney sections of male Wistar rats (control and three NP doses) were stained with hematoxylin–eosin (H&E; 613 images) or immunostained for Beclin-1 (212 images). Whole images yielded 1062 texture features. As animal identity was not recorded, each group’s images were divided into 10 consecutive blocks. Standardization, elastic-net/LASSO selection and classifier training (SVM, KNN, random forest, neural network) used training blocks only (block-grouped 10-fold cross-validation). Results: In H&E images, four-class accuracy was 0.45–0.49 (majority-class rate 0.378). Control versus high dose reached 0.71–0.73 (majority-class rate 0.549), but control versus medium dose (0.66–0.67) stayed below its majority-class rate (0.688). Per block, accuracy was 0.79–0.87 and 0.57–0.60, respectively. In Beclin-1 images, four-class accuracy was 0.52–0.60 (majority-class rate 0.325); per block, 0.92–0.95 for control versus high dose and 0.63–0.73 for control versus medium dose. Conclusions: With block-grouped validation, both stains separated high-dose images from controls, low-dose images in part and medium-dose images least well, a non-monotonic pattern that does not support dose-level classification. Blocks only approximate animals, so generalization to unseen animals is not shown; the control class composition is uncertain and, in H&E images, group is confounded with acquisition order. Findings are preliminary. Confirmation requires animal-labeled images, larger cohorts and pathology scores.