Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Jan 1;16(2):506-520.
doi: 10.7150/jca.101574. eCollection 2025.

Hybrid Model with Wavelet Decomposition and EfficientNet for Accurate Skin Cancer Classification

Amina Aboulmira  1 , Hamid Hrimech  1 , Mohamed Lachgar  2   3 , Mohamed Hanine  4 , Carlos Osorio Garcia  5   6   7 , Gerardo Mendez Mezquita  5   6   7 , Imran Ashraf  8
Affiliations

Hybrid Model with Wavelet Decomposition and EfficientNet for Accurate Skin Cancer Classification

Amina Aboulmira et al. J Cancer. .

Abstract

Faced with anomalies in medical images, Deep learning is facing major challenges in detecting, diagnosing, and classifying the various pathologies that can be treated via medical imaging. The main challenges encountered are mainly due to the imbalance and variability of the data, as well as its complexity. The detection and classification of skin diseases is one such challenge that researchers are trying to overcome, as these anomalies present great variability in terms of appearance, texture, color, and localization, which sometimes makes them difficult to identify accurately and quickly, particularly by doctors, or by the various Deep Learning techniques on offer. In this study, an innovative and robust hybrid architecture is unveiled, underscoring the symbiotic potential of wavelet decomposition in conjunction with EfficientNet models. This approach integrates wavelet transformations with an EfficientNet backbone and incorporates advanced data augmentation, loss function, and optimization strategies. The model tested on the publicly accessible HAM10000 and ISIC2017 datasets has achieved an accuracy rate of 94.7%, and 92.2% respectively.

Keywords: convolutional neural networks; image processing; skin lesion; transfer learning; wavelet decomposition.

PubMed Disclaimer

Conflict of interest statement

Competing Interests: The authors have declared that no competing interest exists.

Figures

Figure 1
Figure 1
Wavelet decomposition levels of a dermoscopic image.
Figure 2
Figure 2
Naive Wavelet convolutional neural network.
Figure 3
Figure 3
General workflow of the proposed hybrid model.
Figure 4
Figure 4
ROC Curve for Wave-EfficientNet Model.
Figure 5
Figure 5
Matrix confusion of the best model performance.
Figure 6
Figure 6
Grad-CAM Visualizations.
See this image and copyright information in PMC

References

    1. Cockburn IM, Henderson R, Stern S. The impact of artificial intelligence on innovation. Natl Bur Econ Res Work Pap Ser. 2018. 24449.
    1. Chen L, Li S, Bai Q. et al. Review of image classification algorithms based on convolutional neural networks. Remote Sens. 2021;13:4712.
    1. Mo Y, Qin H, Dong Y, Zhu Z, Li Z. Large language model (LLM) AI text generation detection based on transformer deep learning algorithm. arXiv. 2024. arXiv:240506652.
    1. Schneider J, Vlachos M. A survey of deep learning: From activations to transformers. Int Conf Agents Artif Intell. 2023.
    1. Castiglioni I. et al. AI applications to medical images: From machine learning to deep learning. Phys Med. 2021;83:9–24. - PubMed

LinkOut - more resources