Handling missing DCE data in prostate cancer detection using multiparametric MRI
Hussam Al-Deen Ashab 1 , Piotr Kozlowski 1 , Robert Rohling 1 , Purang Abolmaesumi 1 , Larry Goldenberg 1 , and Mehdi Moradi 1
University of British Columbia, Vancouver,
The objective of the work presented here is to design
classifiers to detect prostate cancer from MRI
parametric maps with the capability of handling missing
data, specifically DCE parameters. We propose two
different methods and show their effectiveness in
maintaining high AUC while handling missing parameters.
Both methods are based on support vector machine
classification. However, one method trains a single
classifier and uses k-nearest neighbor imputation of DCE
parameters in test cases where DCE is missing. The other
method uses two different classifiers trained on DTI and
DCE, fuses the two methods in cases where both DTI and
DCE are available. We showed that as an increasing
number of cases with missing DCE features are presented
to the classifiers, KNN imputation of missing features
outperforms the fusion of two classifiers. Both methods
outperform a DTI only classifier.
This abstract and the presentation materials are available to members only;
a login is required.