An Efficient Deep Learning-Based Skin Cancer Classifier for an Imbalanced Dataset.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Dermatology

Background:

• Skin cancer detection is time-consuming and challenging, with a shortage of dermatologists globally.
• Early diagnosis is critical for successful skin cancer treatment outcomes.
• Data imbalance in medical datasets hinders the performance of deep learning models.

Purpose of the Study:

• To propose a novel deep learning framework for efficient skin cancer detection using imbalanced image datasets.
• To improve the accuracy and reliability of automated skin cancer diagnosis.
• To overcome data imbalance issues in skin lesion classification.

Main Methods:

• Utilized data augmentation techniques to balance imbalanced skin cancer classes.
• Employed deep learning models including AlexNet, InceptionV3, and RegNetY-320 on the HAM10000 dataset.
• Tuned hyperparameters for optimal performance of the proposed framework.

Main Results:

• The RegNetY-320 model achieved superior performance compared to AlexNet and InceptionV3.
• The proposed framework demonstrated high accuracy (91%), F1-score (88.1%), and ROC curve value (0.95).
• Results significantly surpassed state-of-the-art methods (85% accuracy, 69.3% F1-score, 0.90 ROC).

Conclusions:

• The novel deep learning framework offers a promising solution for efficient and accurate skin cancer detection.
• The method can assist in early disease identification, potentially saving lives and reducing healthcare costs.
• The approach effectively handles imbalanced datasets, enhancing diagnostic capabilities for dermatologists and healthcare professionals.