Brought to you by
ECR 2015 / B-0244
Large-scale objective comparison of 29 novel algorithms for computer-aided diagnosis of dementia based on structural MRI
This poster is published under an open license. Please read the disclaimer for further details.
Congress: ECR 2015
Poster No.: B-0244
Type: Scientific Paper
Keywords: Computer applications, Neuroradiology brain, MR, Computer Applications-Detection, diagnosis, Comparative studies, CAD, Dementia
Authors: E. Bron1, M. Smits1, F. Barkhof2, A. J. Bastos-Leite3, J. C. van Swieten1, W. J. Niessen1, S. Klein1; 1Rotterdam/NL, 2Amsterdam/NL, 3Porto/PT
DOI:10.1594/ecr2015/B-0244

Results

As shown in Fig. 4 and Fig. 5, the algorithms used a wide range of approaches. Different feature extraction methods and classifiers were used. Most of the algorithms were trained on the 30 provided training sets and on data from ADNI and AIBL. 

 

Fig. 4: Overview of the types of features used by the 29 evaluated algorithms.
References: Radiology and Medical Informatics, Erasmus MC - Rotterdam/NL

 

Fig. 5: Overview of the types of classifiers used by the 29 evaluated algorithms.
References: Radiology and Medical Informatics, Erasmus MC - Rotterdam/NL

  

The performances for the five best performing methods are shown in Fig. 6 (accuracy) and Fig. 7 (AUC). Results for the other methods can be found on http://caddementia.grand-challenge.org.

 

Fig. 6: This table presents the classification accuracy of the 5 best-ranked algorithms. TPF = true positive fraction. As three-class classification of AD, MCI and controls is evaluated, the accuracy for random guessing would be ~33%. The full results table is published on http://caddementia.grand-challenge.org
References: Radiology and Medical Informatics, Erasmus MC - Rotterdam/NL

 

Fig. 7: This table presents the area under the receiver operating characteristic (ROC) curve (AUC) of the 5 best-ranked algorithms. The full results table is published on http://caddementia.grand-challenge.org
References: Radiology and Medical Informatics, Erasmus MC - Rotterdam/NL

 

The best performing algorithm (Sørensen et al., 2014) yielded an AUC of 78.8% (CI: 75.6-82.0%) and an accuracy of 63.0% (CI: 57.9-67.5%), which was significantly better than 24 other algorithms. In general, the best performances were achieved using a combination of features that included volume, cortical thickness, shape and intensity.

 

 

POSTER ACTIONS Add bookmark Contact presenter Send to a friend Download pdf
SHARE THIS POSTER
2 clicks for more privacy: On the first click the button will be activated and you can then share the poster with a second click.

This website uses cookies. Learn more