Comparing Clustering Methods for Skin Lesion Detection
Author Information
Author(s): Kockara Sinan, Mete Mutlu, Chen Bernard, Aydin Kemal
Primary Institution: University of Central Arkansas
Hypothesis
This study compares the effectiveness of density-based clustering and fuzzy c-means clustering for automatic border detection in dermoscopy images.
Conclusion
DBSCAN outperformed FCM in accurately delineating skin lesions with lower border error and higher precision and recall.
Supporting Evidence
- DBSCAN achieved an overall average border error ratio of 6.94%.
- FCM had an overall average border error ratio of 100%.
- DBSCAN showed higher precision and recall compared to FCM.
Takeaway
The study tested two computer programs to help doctors find the edges of skin spots. One program did a much better job than the other.
Methodology
The study used 100 dermoscopy images and compared the performance of DBSCAN and FCM clustering algorithms based on border error, precision, and recall metrics.
Limitations
The FCM method showed poor performance, especially in border error metrics, and may over-segment images if the number of clusters is not chosen correctly.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website