STATISTICAL FEATURE LEARNING THROUGH ENHANCED DELAUNAY CLUSTERING AND ENSEMBLE CLASSIFIERS FOR SKIN LESION SEGMENTATION AND CLASSIFICATION
There are multiple types of skin cancer but melanoma is the deadliest skin cancer or lesion type. Early recognition of melanoma in dermoscopy images essentially increase the endurance rate. However, the precise acknowledgment of melanoma is very challenging because of the numerous reasons: low diffe...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Little Lion Scientific
2021
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/47356/1/11Vol99No5.pdf http://ir.unimas.my/id/eprint/47356/ https://www.jatit.org/volumes/Vol99No5/11Vol99No5.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | There are multiple types of skin cancer but melanoma is the deadliest skin cancer or lesion type. Early recognition of melanoma in dermoscopy images essentially increase the endurance rate. However, the precise acknowledgment of melanoma is very challenging because of the numerous reasons: low difference among lesion and skin, visual comparability among melanoma and non-melanoma lesions, and so forth. Consequently, the dependable programmed diagnosis of skin cancer is exceptionally helpful to dermatologist.
In this paper, we proposed profound learning strategy to address three primary assignments developing in the
zone of skin lesion picture preparation, i.e., dermoscopic highlight, extraction and detection. A profound algorithm comprising of preprocessing in CIELAB color space and Delaunay triangulation based clustering along with Particle Swarm Optimization (PSO) is proposed for the segmentation. Moreover, skin lesion images are clustered based on fused color, pattern and shape based features. A boost ensemble learning algorithm using Support Vector Machines (SVM) as initial classifiers and Artificial Neural Networks (ANN)
as a final classifier is employed to learn the patterns of different skin lesion class features. The proposed
automated system is assessed on the ISIC and PH2 datasets. Test results show the promising efficiency of our proposed study, i.e., 96.8% and 92.1% segmentation accuracy for ISIC and PH2 datasets respectively. Classification accuracy of 97.9% also accomplished on ISIC dataset. It can be concluded from this research that proposed system employed the power of simple methods that are less resource-hungry yet provide better results. |
|---|