Using Bayesian neural networks with ARD input selection to detect malignant ovarian masses prior to surgery

Authors:
Ben Van Calster;Dirk Timmerman;Ian T. Nabney;Lil Valentin;Antonia C. Testa;Caroline Van Holsbeke;Ignace Vergote;Sabine Van Huffel
Affiliations:
Katholieke Universiteit Leuven, Department of Electrical Engineering (ESAT-SCD), Kasteelpark Arenberg 10, 3001, Leuven, Belgium;University Hospitals Leuven, Department of Obstetrics and Gynaecology, Herestraat 49, 3000, Leuven, Belgium;Aston University, Neural Computing Research Group (NCRG), Aston Triangle, B4 7ET, Birmingham, UK;Malmö University Hospital, Department of Obstetrics and Gynaecology, Aston Triangle, 205 02, Malmö, Sweden;Università Cattolica del Sacro Cuore, Gynecologic Oncology Unit, Largo Agostino Gemelli 8, 00168, Rome, Italy;University Hospitals Leuven, Department of Obstetrics and Gynaecology, Herestraat 49, 3000, Leuven, Belgium;University Hospitals Leuven, Department of Obstetrics and Gynaecology, Herestraat 49, 3000, Leuven, Belgium;Katholieke Universiteit Leuven, Department of Electrical Engineering (ESAT-SCD), Kasteelpark Arenberg 10, 3001, Leuven, Belgium
Venue:
Neural Computing and Applications
Year:
2008

Citing 0
Cited 2

An application of methods for the probabilistic three-class classification of pregnancies of unknown location

Artificial Intelligence in Medicine
Clinical decision support for ovarian tumor diagnosis using Bayesian models: Results from the IOTA study

Proceedings of the 2009 conference on Computational Intelligence and Bioengineering: Essays in Memory of Antonina Starita

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, we applied Bayesian multi-layer perceptrons (MLP) using the evidence procedure to predict malignancy of ovarian masses in a large (n = 1,066) multi-centre data set. Automatic relevance determination (ARD) was used to select the most relevant inputs. Fivefold cross-validation (5CV) and repeated 5CV was used to select the optimal combination of input set and number of hidden neurons. Results indicate good performance of the models with area under the receiver operating characteristic curve values of 0.93–0.94 on independent data. Comparison with a linear benchmark model and a previously developed logistic regression model shows that the present problem is very well linearly separable. A resampling analysis further shows that the number of hidden neurons specified in the ARD analyses for input selection may influence model performance. This paper shows that Bayesian MLPs, although not frequently used, are a useful tool for detecting malignant ovarian tumours.