Related Experiment Video
Updated: Jul 15, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Preoperative diagnosis of ovarian tumors using Bayesian kernel-based methods.
B Van Calster1, D Timmerman, C Lu
1Department of Electrical Engineering (ESAT-SCD), Katholieke Universiteit Leuven, and Department of Obstetrics and Gynecology, University Hospitals K. U. Leuven, Belgium. ben.vancalster@esat.kuleuven.be
Summary
Bayesian kernel-based machine learning accurately distinguishes malignant from benign adnexal masses. These flexible classifiers achieved high sensitivity and specificity, aiding in diagnosing ovarian cancer.
Area of Science:
- Oncology
- Medical Imaging
- Machine Learning
Background:
- Adnexal masses require accurate malignancy prediction for timely treatment.
- Current diagnostic methods can be limited in distinguishing benign from malignant cases.
Purpose of the Study:
- To develop and evaluate flexible machine learning classifiers for predicting adnexal mass malignancy.
- To compare Bayesian least squares support vector machines (BLSSVM) with relevance vector machines (RVM).
Main Methods:
- Utilized a large database of 1066 patients with adnexal masses from nine centers.
- Developed and tested models using clinical and ultrasound data, with histological classification as the outcome.
- Employed Bayesian least squares support vector machines and relevance vector machines for prediction.
Main Results:
- Twenty-five percent of masses were malignant. A set of 12 key variables predicted malignancy.
- Models achieved an area under the receiver-operating characteristics curve (AUC) exceeding 0.940.
- The BLSSVM with a linear kernel showed an AUC of 0.946, 91% sensitivity, and 84% specificity.
Conclusions:
- Bayesian kernel-based methods effectively differentiate malignant from benign adnexal masses.
- The developed models demonstrate robustness across multiple centers.
- Further studies will investigate the long-term robustness of these predictive models.
