Fri, Aug 14, 2026
[Archive]
Volume 18, Issue 2 (June-2026 2026)
Iranian Journal of Blood and Cancer 2026, 18(2): 15-24 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Heyrani Khameneh A, Valizadeh M, Chehel Amirani M, Kashef S S. HDLF-CBHIR: A Hierarchical Deep Learning Framework for Content-Based Histopathological Images Retrieval. Iranian Journal of Blood and Cancer 2026; 18 (2) :15-24
URL: http://ijbc.ir/article-1-1887-en.html
URL: http://ijbc.ir/article-1-1887-en.html
1- Department of Electrical and Computer Engineering, Urmia University, Urmia, Iran.
2- Department of Electrical and Computer Engineering, Urmia University, Urmia, Iran. ,mo.valizadeh@urmia.ac.ir
2- Department of Electrical and Computer Engineering, Urmia University, Urmia, Iran. ,
Abstract: (12 Views)
Objective: To propose a deep learning-based hierarchical framework for content-based histopathology image retrieval (CBHIR) that can serve as a diagnostic aid for pathologists.
Methodology: The framework extracts deep features using a pre-trained EfficientNet-B0 model, followed by global average pooling (GAP) for dimensionality reduction. A triplet network with a custom loss function is designed for deep hashing; the loss function leverages cosine similarity and incorporates data engineering by selecting the most challenging samples in the feature space to enhance retrieval performance. The model is evaluated on the Kimia path24c dataset and compared with baseline methods.
Results: The proposed framework achieved an accuracy of 98.76% for patch-level retrieval, 98.61% for whole-slide retrieval, and a total accuracy of 97.39% in publicly available Kimia path24c dataset.
Conclusion: The experimental results demonstrate the efficiency of the proposed framework for the CBHIR task. Moreover, the output of this work can be used in diagnostic practice to help pathologists analyze digital histopathology samples, as well as in education to support training of pathology professionals.
Methodology: The framework extracts deep features using a pre-trained EfficientNet-B0 model, followed by global average pooling (GAP) for dimensionality reduction. A triplet network with a custom loss function is designed for deep hashing; the loss function leverages cosine similarity and incorporates data engineering by selecting the most challenging samples in the feature space to enhance retrieval performance. The model is evaluated on the Kimia path24c dataset and compared with baseline methods.
Results: The proposed framework achieved an accuracy of 98.76% for patch-level retrieval, 98.61% for whole-slide retrieval, and a total accuracy of 97.39% in publicly available Kimia path24c dataset.
Conclusion: The experimental results demonstrate the efficiency of the proposed framework for the CBHIR task. Moreover, the output of this work can be used in diagnostic practice to help pathologists analyze digital histopathology samples, as well as in education to support training of pathology professionals.
Keywords: Artificial intelligence (AI), Deep learning, Content-based image retrieval (CBIR), Digital pathology, Triplet network, Diagnosis, Cosine similarity
: Original Article |
Subject:
AI in Medicine
Received: 2026/04/22 | Accepted: 2026/06/8 | Published: 2026/06/30
Received: 2026/04/22 | Accepted: 2026/06/8 | Published: 2026/06/30
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
- Iranian Journal of Blood and Cancer
- Journal Tel: +989369936116
- Publisher Tel: +982122275917
- Mail: Info@ijbc.ir
