Paper Info
Title
Unbiased confidence measures for stroke risk estimation based on ultrasound carotid image analysis.
Abstract
We propose an approach for providing well-calibrated confidence measures for determining cerebrovascular risk stratification based on characteristics from noninvasive ultrasound imaging of carotid plaques. An important challenge we address is the class imbalance problem inherent in the particular task. The proposed approach is based on a novel framework, called conformal prediction (CP), for developing techniques that output sets of predictions guaranteed to contain the true classification of a new case with a prespecified probability. We follow a modified version of the CP framework, called Label-conditional Mondrian conformal prediction (LCMCP), so that the guarantee provided by CP does not only hold for all instances together, but also hold for the instances of each class independently, thus making prediction sets unbiased. Furthermore, LCMCP is combined with an underbagging ensemble of artificial neural networks so that its outputs are based on unbiased estimates. The important positive properties of the proposed approach are demonstrated experimentally on a dataset of patients that were followed up for eight years and had asymptomatic internal carotid artery stenosis at the baseline.
Year
DOI
Venue
2017
10.1007/s00521-016-2590-3
Neural Computing and Applications
Keywords
Field
DocType
Conformal prediction, Class imbalance, Confidence measures, Stroke risk assessment, Plaque imaging, Computer-aided diagnosis, Ultrasound image analysis
Internal carotid artery stenosis,Confidence measures,Computer-aided diagnosis,Ultrasound imaging,Stroke,Artificial intelligence,Artificial neural network,Mathematics,Machine learning,Ultrasound
Journal
Volume
Issue
ISSN
28
6
1433-3058
Citations 
PageRank 
References 
2
0.39
19
Authors
3
Name
Order
Citations
PageRank
Harris Papadopoulos121926.33
Efthyvoulos Kyriacou2308.21
A. Nicolaides3727.46