Automatic skin lesions detection from images through microscopic hybrid features set and machine learning...

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Skin cancer incidence is rising globally, highlighting the need for improved diagnostic tools.
• Manual diagnosis of skin lesions can be subjective and vary in accuracy.
• Automated systems can aid physicians in precise, early-stage skin cancer detection, potentially reducing mortality rates.

Purpose of the Study:

• To develop an automated system for accurate skin lesion detection and classification.
• To leverage deep learning and machine learning for enhanced diagnostic precision.

Main Methods:

• Utilized deep features extracted via the AlexNet architecture combined with a local optimal-oriented pattern.
• Employed a hybrid feature set for machine learning-based classification.
• Tested the model on the PAD-UFES-20 and MED-NODE datasets containing melanoma and nevus images.

Main Results:

• The proposed model demonstrated accurate prediction of skin lesions using deep features.
• Experimental results on both datasets confirmed the efficacy of hybrid features and machine learning.
• Achieved a classification accuracy of 94.7% using an ensemble classifier.

Conclusions:

• The developed automated system effectively detects skin lesions with high accuracy.
• The integration of deep features and machine learning shows significant promise for early skin cancer diagnosis.