Melanoma Detection Using Deep Learning-Based Classifications

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin cancer is a prevalent global health concern, with early diagnosis crucial for effective treatment.
• Deep learning (DL) shows promise for automated diagnostic systems in healthcare.
• Existing methods for skin lesion analysis can be improved for accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a DL-based automated system for precise skin lesion segmentation and classification.
• To enhance medical professionals' capabilities in diagnosing various types of skin cancer.
• To improve the accuracy and efficiency of skin cancer diagnosis through advanced image analysis.

Main Methods:

• Image enhancement using Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN).
• Segmentation of Regions of Interest (ROI) for lesion localization.
• Data augmentation to address data imbalance.
• Classification of skin lesions using a Convolutional Neural Network (CNN) and a modified Resnet-50 model on the HAM10000 dataset.

Main Results:

• The proposed CNN-based model achieved an accuracy of 0.86, precision of 0.84, recall of 0.86, and an F-score of 0.86.
• The automated system demonstrated superior performance compared to previous studies.
• The method effectively segmented lesion zones and classified seven types of skin cancer.

Conclusions:

• The developed DL model offers an improved automated approach for skin cancer diagnosis.
• This system can aid medical professionals in making more accurate and timely diagnoses.
• The enhanced automated method has the potential to benefit both healthcare providers and patients through earlier and more precise detection.