EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Explainable Hybrid Deep‐Radiomic Graph Learning Framework for Automated IDC Aggressiveness Grading From Whole‐Slide Histopathology Images

Vijayalaxmi Inamdar iD, Shaila S G iD

International Journal of Biomedical Imaging · 2026-01 · 2026 (1)

DOI: 10.1155/ijbi/7498792 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Biomedical Imaging, online erschienen am 10.11.2026. Verfasst von Vijayalaxmi Inamdar und 1 weiteren Autor:innen. Beteiligte Einrichtung: Dayananda Sagar Academy of Technology and Management Bengaluru Karnataka India dayanandasagar.edu. Die Arbeit nennt 53 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Breast cancer remains one of the leading causes of cancer‐related mortality among women worldwide. Histopathological analysis of whole‐slide images (WSIs) plays a crucial role in the diagnosis and grading of breast cancer subtypes. Invasive ductal carcinoma (IDC) is the most prevalent and heterogeneous subtype of breast cancer, and its aggressiveness grading has important prognostic value. Manual grading of IDC aggressiveness using hematoxylin and eosin (H&E) stained WSIs is tedious and subjective. In this study, we address the issues of automated IDC aggressiveness grading using an explainable hybrid deep‐radiomic graph learning framework. The proposed pipeline jointly performs tumor segmentation, feature representation, and classification within an end‐to‐end framework. The tumor is segmented using an attention U‐Net followed by dense conditional random field (DenseCRF) segmentation refinement. Deep convolutional features and radiomic features encoding texture, morphology, and intensity statistics are extracted from the segmented tumor region and combined to form hybrid deep‐radiomic features. Spatial graphs constructed from these hybrid features are subsequently processed using a graph attention network (GAT)–based graph neural network (GNN) with attention‐based multiple instance learning (MIL) to reason about the relationships between different patches for slide‐level grade prediction. Extensive experiments performed on two datasets, BreakHis and BACH, show that our proposed architecture demonstrates improved performance in IDC aggressiveness grading when compared with convolutional neural network (CNN) and GNN baselines with an accuracy of 98.3%, macro‐F1 score of 0.968, ROC‐AUC score of 0.98, and Dice score of 0.932. The explainability of proposed model with Grad‐CAM visual explanations, graph attention scores, and SHAP‐based feature importance scores for individual radiomic features are presented.

Schlagwörter

Stichwörter der Autor:innen

breast cancerdeep-radiomic modeldense conditional random fieldgraph attention networkgraph neural networkhistopathologyinvasive ductal carcinomamultiple instance learning

Autor:innen und Institutionen

  • Dayananda Sagar Academy of Technology and Management Bengaluru Karnataka India dayanandasagar.edu.
  • School of Engineering Dayananda Sagar University Bengaluru Karnataka India.

ORCID-Profile: Vijayalaxmi Inamdar, Shaila S G

Zitieren

APA

Inamdar, V., & S G, S. (2026). Explainable Hybrid Deep‐Radiomic Graph Learning Framework for Automated IDC Aggressiveness Grading From Whole‐Slide Histopathology Images. International Journal of Biomedical Imaging, 2026(1). https://doi.org/10.1155/ijbi/7498792

Vancouver

Inamdar V, S G S. Explainable Hybrid Deep‐Radiomic Graph Learning Framework for Automated IDC Aggressiveness Grading From Whole‐Slide Histopathology Images. International Journal of Biomedical Imaging. 2026;2026(1). doi:10.1155/ijbi/7498792

BibTeX
@article{Inamdar2026,
  author = {Inamdar, Vijayalaxmi and S G, Shaila},
  title = {Explainable Hybrid Deep‐Radiomic Graph Learning Framework for Automated IDC Aggressiveness Grading From Whole‐Slide Histopathology Images},
  journal = {International Journal of Biomedical Imaging},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/ijbi/7498792}
}
Kostenlos starten →