Related Experiment Video
Updated: Aug 6, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
CT-Based Radiomic Features Predict Cervical Lymph Node Metastasis in Dogs With Oral Malignancy: A Machine Learning
Christopher J Pinard1,2,3,4, Andrew Lagree2, Ryan Appleby5
1Department of Oncology, Toronto Animal Cancer Centre, Toronto, Ontario, Canada.
Veterinary and Comparative Oncology
|July 21, 2026
Summary
Accurate identification of canine cervical lymph node metastasis using CT radiomics shows modest performance. Texture features offer a reproducible signal, but external validation is needed for clinical use.
Area of Science:
- Veterinary Radiology
- Medical Imaging
- Machine Learning in Oncology
Background:
- Accurate preoperative identification of cervical lymph node (LN) metastasis is crucial for staging and treatment planning in dogs with oral malignancy.
- Conventional imaging methods have limited diagnostic sensitivity for detecting LN metastasis in dogs.
Purpose of the Study:
- To evaluate the efficacy of CT-derived radiomic features combined with machine learning classifiers in predicting the metastatic status of mandibular and retropharyngeal LNs in dogs with oral malignancy.
- To assess the reproducibility and biological interpretability of CT-derived texture features for cervical LN metastasis detection.
Main Methods:
- Retrospective study of 49 dogs with histopathologically confirmed oral malignancy undergoing contrast-enhanced CT.
- Manual segmentation of four bilateral cervical LN sites, extraction of 107 radiomic features using PyRadiomics.
- Feature selection using variance filtering, Spearman redundancy removal, and Mann-Whitney U testing with Benjamini-Hochberg correction within leave-one-patient-out cross-validation (LOOCV).
- Training of logistic regression, random forest (RF), support vector machine, and XGBoost classifiers with SMOTE applied within training folds.
- Evaluation of per-LN and per-patient performance using out-of-fold predictions, with confidence intervals derived from patient-level bootstrap.
Main Results:
- At the per-LN level, RF achieved an AUC of 0.649 and XGBoost an AUC of 0.631.
- Patient-level aggregation yielded an RF AUC of 0.613, identifying 9 out of 13 metastatic-positive dogs.
- Four GLSZM/GLCM texture features were consistently selected and remained significant under generalized estimating equation (GEE) inference, indicating a reproducible signal.
Conclusions:
- CT-derived texture features possess a reproducible and biologically interpretable signal for canine cervical LN metastasis.
- The leakage-controlled performance of radiomic features and machine learning classifiers is modest in this proof-of-concept study.
- External validation in larger, multi-institutional cohorts is essential before clinical translation of this approach.
