QuRad: Radiomics Feature Extraction for QuPath
- 1Sorbonne Université, Institut du Cerveau-Paris Brain Institute-ICM, CNRS, Inria, Inserm, AP-HP, Hôpital de la Pitié Salpêtrière, Paris, France.
- 2Centre for Eye Research Australia, University of Melbourne, Royal Victorian Eye and Ear Hospital, East Melbourne, Victoria, Australia.
- 3LIX, École Polytechnique, Palaiseau, France.
Abstract
Radiomic features provide a quantitative description of tissue organization in histopathology images. QuPath, a widely used open-source platform for whole-slide image analysis, offers morphology, intensity, and Haralick texture measurements but no standardized radiomic feature set such as that of PyRadiomics. Researchers must therefore export data to external Python pipelines, interrupting their QuPath workflow. QuRad extracts radiomic features from cell detections and annotation regions directly inside QuPath. The extension and equivalent Groovy script implement 119 features: 103 two-dimensional features enabled by default and 16 optional legacy shape descriptors. We compared QuRad 0.4 with PyRadiomics 3.0.1 on identical grayscale images and pixel masks. All 103 default features met a 10 -9 absolute-or-relative agreement tolerance on a breast-cancer benchmark tile (410 cells). On 20 melanoma tiles (8,644 nuclei), 97 of 103 features met this tolerance and 6 polygon-derived shape features differed from PyRadiomics' raster-derived mesh (minimum concordance correlation coefficient 0.942). The 22 synthetic regions accepted by PyRadiomics reproduced the expected intensity and texture values. A single-threaded server benchmark reached about 1,900 nuclei per second, and processing the 184,242 cells of a whole-slide image took 2.7 min on a laptop. A random forest distinguished tumor cells from lymphocytes in melanoma (pooled macro-F1 0.853) and classified 197 tissue regions of 6 breast-cancer slides into 3 compartments (0.696). QuRad exports the features with object coordinates and extraction settings for use in QuPath or downstream analyses.
Related
- Exploring self-supervised deep sparse autoencoders for robust feature selection in radiomics analysis
- Hybrid feature pyramid attention and capsule network (FpaNet+CapsNet) for enhanced spatial and contextual feature extraction for improved brain tumor segmentation
- Multi-task deep learning framework for brain tumor classification and segmentation using ensemble feature extraction
- Comparative Evaluation of Feature Selection Algorithms for Predicting Radiation-Induced Complications in Head and Neck Cancer Using Multimodal CT and MRI Radiomics
- Enhancing medical image segmentation with HCE-Net: A heterogeneous feature collaborative extraction network
BibTeX
@article{arslan2026qurad,
title = {{QuRad: Radiomics Feature Extraction for QuPath}},
author = {Janan Arslan and Tamara Künzle and Yeşim Güner and Agustí Alentorn},
journal = {Computational and Structural Biotechnology Journal},
year = {2026},
month = oct,
doi = {10.34133/csbj.0238},
url = {https://hal.science/hal-05820407v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - QuRad: Radiomics Feature Extraction for QuPath
AU - Janan Arslan
AU - Tamara Künzle
AU - Yeşim Güner
AU - Agustí Alentorn
PY - 2026
DA - 2026/10/19
JO - Computational and Structural Biotechnology Journal
DO - 10.34133/csbj.0238
UR - https://hal.science/hal-05820407v1
AB - Radiomic features provide a quantitative description of tissue organization in histopathology images. QuPath, a widely used open-source platform for whole-slide image analysis, offers morphology, intensity, and Haralick texture measurements but no standardized radiomic feature set such as that of PyRadiomics. Researchers must therefore export data to external Python pipelines, interrupting their QuPath workflow. QuRad extracts radiomic features from cell detections and annotation regions directly inside QuPath. The extension and equivalent Groovy script implement 119 features: 103 two-dimensional features enabled by default and 16 optional legacy shape descriptors. We compared QuRad 0.4 with PyRadiomics 3.0.1 on identical grayscale images and pixel masks. All 103 default features met a 10 -9 absolute-or-relative agreement tolerance on a breast-cancer benchmark tile (410 cells). On 20 melanoma tiles (8,644 nuclei), 97 of 103 features met this tolerance and 6 polygon-derived shape features differed from PyRadiomics' raster-derived mesh (minimum concordance correlation coefficient 0.942). The 22 synthetic regions accepted by PyRadiomics reproduced the expected intensity and texture values. A single-threaded server benchmark reached about 1,900 nuclei per second, and processing the 184,242 cells of a whole-slide image took 2.7 min on a laptop. A random forest distinguished tumor cells from lymphocytes in melanoma (pooled macro-F1 0.853) and classified 197 tissue regions of 6 breast-cancer slides into 3 compartments (0.696). QuRad exports the features with object coordinates and extraction settings for use in QuPath or downstream analyses.
SN - 2001-0370
ER -