A multi-level closing based segmentation framework for dermatoscopic images using ensemble deep network

Skin cancer, especially melanoma is a lethal form of cancer whose prevalence is increasing in recent times with increased exposure to ultra-violet rays and use of harmful skin cosmetics. The proposed methodology aims at providing a highly optimised pedagogy for lesion segmentation in dermatoscopic images. It is a hybrid model with an extensive pre-processing for hair removal by applying multi-level closing operation followed by segmentation using an ensemble deep network. Two publicly available datasets viz. HAM10K and ISIC 2018 are used to analyse the performance of the framework. The average values of Dice Coefficient and Jaccard value for both datasets are found to be 0.9555 and 0.8545 respectively. Also, the proposed framework achieved an average accuracy of 95.87% for both datasets which outperformed all base models and also the proposed framework without pre-processing.