Figure 2.

Performance comparison with K-means clustering, Otsu clustering, the proposed entropy-based method. (a) an IHC Lung carcinoma tissue image, (b) poor segmentation by Otsu unsupervised clustering, which automatically separates the image into two classes but contains a lot of false detection, (c)(d) poor segmentation by K-means clustering, which automatically separates the image into three classes here but the resulting clusters are poor in nuclear segmentation, (e) over-segmentation by Vincent-Soille watershed algorithm [12] (f) poor segmentation result by marker-controlled watershed method [14] (g) segmentation result with many false positives by optimized watershed transformation (adapted from [13]) (h) improved nuclear segmentation by the proposed entropy-based method.

Wang BMC Bioinformatics 2012 13:21   doi:10.1186/1471-2105-13-21
Download authors' original image