Author(s):
1. Muhammad Hamza Mir:
Department of Computer Science, Faculty of Telecommunication and Information Engineering, University of Engineering and Technology,Taxila, Pakistan
2. Muhammad Javed Iqbal:
Department of Computer Science, Faculty of Telecommunication and Information Engineering, University of Engineering and Technology,Taxila, Pakistan
3. Syed Aun Irtaza:
Department of Computer Science, Faculty of Telecommunication and Information Engineering, University of Engineering and Technology,Taxila, Pakistan
Abstract:
Skin cancer is one of the most lethal types of cancer accounting for a large number of deaths worldwide. Hence, correct and timely diagnosis is crucial since skin cancer has a high fatality rate and slowly spreads to other regions of the body. However, it is exceedingly difficult to diagnose skin cancer through eye inspection and manual examination of skin lesions. Manual inspection may lead to misinterpretation in cancer boundaries and type due to similarity between lesions. Hence, the researchers have automated systems using Machine Learning (ML). Despite tremendous progress, the conventional ML based methodologies rely on hand-engineered feature extraction which require human intervention and selection of algorithm to extract optimal features. In comparison, the Deep Learning (DL) based systems are completely automated and produce very robust and accurate results. We propose a completely automated skin lesion segmentation and classification method using deep learning. The presented segmentation framework makes use of encoder-decoder based UNET architecture. Whereas, the features are extracted using famous DL frameworks i.e. ResNet50, DenseNet201, AlexNet8 and VGG16. The features from highest performing classifiers are combined to form a single vector which are then classified via SVM and KNN. The segmentation achieved highest accuracy of 95% through UNET and the classification framework successfully classified benign and malignant lesions with 87% accuracy by using SVM. The suggested framework is capable of real-time lesion segmentation and classification of skin lesions from dermoscopic images.
Page(s):
1-1
DOI:
DOI not available
Published:
Journal: Second International Conference on Computing Technologies, Tools and Applications (ICTAPP-24), June 4-6,2024 (Abstract Book), Volume: 0, Issue: 0, Year: 2024
Keywords:
deep learning
,
Melanoma Segmentation
,
Artificial Intelligence
,
Melanoma Classification
References:
References are not available for this document.
Citations
Citations are not available for this document.