Hybrid Model for Skin Cancer Classification
Author Information
Author(s): Aboulmira Amina, Hrimech Hamid, Lachgar Mohamed, Hanine Mohamed, Garcia Carlos Osorio, Mezquita Gerardo Mendez, Ashraf Imran
Primary Institution: LAMSAD Laboratory, ENSA, Hassan First University, Berrechid, Morocco.
Hypothesis
Can a hybrid model combining wavelet decomposition and EfficientNet improve skin cancer classification accuracy?
Conclusion
The hybrid model achieved an accuracy of 94.7% on the HAM10000 dataset and 92.2% on the ISIC2017 dataset.
Supporting Evidence
- The model achieved an accuracy rate of 94.7% on the HAM10000 dataset.
- The model achieved an accuracy rate of 92.2% on the ISIC2017 dataset.
- The hybrid model combines wavelet decomposition with EfficientNet for improved feature extraction.
- Data augmentation techniques were used to enhance model generalization.
- The model was tested on two well-established datasets for skin disease classification.
Takeaway
This study created a new way to help computers recognize skin cancer better by using special math and smart computer models.
Methodology
The study used a hybrid CNN architecture that integrates wavelet-transformed inputs into an EfficientNet backbone for skin lesion classification.
Potential Biases
The model may be biased towards more common skin conditions due to class imbalance in the training data.
Limitations
The model's performance may vary with image quality and resolution, and it may struggle with rare skin conditions.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website