Skip to main content
. Author manuscript; available in PMC: 2014 Jun 1.
Published in final edited form as: J Biomed Inform. 2013 Feb 27;46(3):436–443. doi: 10.1016/j.jbi.2013.02.001

Table 3.

Results of the different clustering methods for the Pubmed images data.

Method descriptor Clustering algorithm Data modality
utilized		 Micro-Averaged Precision* (95%
confidence interval)		 NMI* (95%
confidence
interval)
KmeansHaralick Kmeans Haralick 0.338 (0.332, 0.343) 0.136 (0.135, 0.138)
KmeansCaptions Kmeans Captions 0.433 (0.43, 0.44) 0.182 (0.177, 0.186)
EnsembleHaralick Ensemble Haralick 0.340 (0.335, 0.346) 0.154 (0.153, 0.155)
EnsembleCaptions Ensemble Captions 0.465 (0.460, 0.470) 0.179 (0.177, 0.182)
EnsembleMerged Ensemble of merged modalities Haralick and captions merged 0.360 (0.351, 0.370) 0.134 (0.131, 0.137)
Complementary Ensemble Clustering Linear combination of coassociation matrices from ensemble clustering of each modality separately Haralick and captions separately 0.474 (0.468, 0.480) 0.189 (0.186, 0.192)

NMI: Normalized Mutual Information.

Bold font indicates highest performing method(s).

*

Larger is better.