In an article recently published in Engineering Science and Technology, an International Journal, researchers introduced a novel artificial intelligence (AI) technique for skin cancer prediction. Their aim was to combine deep learning (DL) with a metaheuristic optimization technique to address the critical need for accurate early detection and classification of various types of skin cancer.
Study: Boosting Skin Cancer Detection with AI Techniques. Image Credit: PopTika/Shutterstock.com
Background
Skin cancer is a major public health concern affecting millions worldwide, regardless of age or skin type. Melanoma, or malignant melanoma, is a particularly dangerous form of skin cancer. It is primarily caused by prolonged exposure to ultraviolet (UV) radiation, which leads to mutations in melanocytes, the cells responsible for producing melanin.
These mutations cause melanocytes to produce excessive amounts of melanin that results in dark moles on the skin. Over time, these lesions or moles may develop into cancerous tumors and spread to other parts of the body.
Traditional diagnostic methods, such as the asymmetry-border-color-diameter (ABCD) rule and histopathological examination, have limitations, including the need for highly experienced dermatologists and potential diagnostic discrepancies. Recent advancements in computer vision and DL, particularly with convolutional neural networks (CNNs), have shown promise in improving automated skin cancer diagnosis.
About the Research
In this paper, the authors proposed a hybrid AI framework that combines the Xception model, a state-of-the-art deep CNN, with the artificial bee colony (ABC) optimization algorithm to improve the accuracy of skin cancer prediction.
The framework consists of two main steps: creating a comprehensive skin cancer dataset and optimizing DL models using the ABC algorithm. The ABC algorithm, inspired by honeybees' foraging behavior, is a metaheuristic optimization technique known for its effectiveness in solving complex optimization problems.
To enhance CNN's robustness, the researchers combined multiple publicly available datasets into a generalized skin cancer dataset. This dataset includes 27,108 images from various sources covering multiple skin cancer types. This diverse dataset supports the training and evaluation of the hybrid model.
The ABC algorithm optimizes CNN training by fine-tuning initial weights and addressing local optima issues. Furthermore, the study evaluated the hybrid framework against eight prominent CNN models, including DenseNet121 and InceptionResNetV2, demonstrating its superior predictive capabilities.
Research Findings
The outcomes demonstrated that the optimized hybrid framework achieved impressive results: an accuracy of 93.04%, recall of 92.0%, precision of 93.0%, and a harmonic mean of precision and recall (F1-score) of 93.12%. This performance significantly outperformed the Xception model without ABC optimization, which had an accuracy of 89.7%.
These results highlight the effectiveness of the ABC algorithm in enhancing DL models for skin cancer classification. The model also maintained high precision, recall, and F1-score across all nine skin cancer classes, demonstrating its robustness in accurately classifying various types of skin cancer.
The authors showed that combining the Xception model feature extraction capabilities with the ABC algorithm's weight optimization prevented the model from getting stuck in local optima. They also emphasized the importance of segmenting skin lesions to improve prediction accuracy. This allows AI models to focus on relevant areas and ignore normal surrounding regions, thereby enhancing the learning process and weight optimization.
Applications
The proposed hybrid framework has the potential to transform skin cancer diagnosis and management significantly. By providing a highly accurate and efficient tool for classifying various types of skin cancer, the model can assist healthcare professionals with early detection and diagnosis.
This can lead to better patient outcomes, lower healthcare costs, and potentially save lives. Additionally, the developed skin cancer dataset can serve as a valuable resource for the research community, supporting the development and evaluation of advanced skin cancer detection and classification algorithms.
Conclusion
In summary, the novel technique proved effective for early detection and accurate classification of skin cancer. The researchers demonstrated that advancements in AI have the potential to revolutionize dermatology and improve patient outcomes. They also acknowledged limitations and challenges, including the need for further validation on larger and more diverse datasets and addressing potential training data biases.
Future work should focus on expanding the generalized skin cancer dataset, exploring ensemble techniques to enhance model robustness, and integrating the framework into clinical decision support systems.
Journal Reference
Farea, E., & et, al. A hybrid deep learning skin cancer prediction framework. Engineering Science and Technology, an International Journal, 57, 101818, 2024. DOI: 10.1016/j.jestch.2024.101818, https://www.sciencedirect.com/science/article/pii/S2215098624002040
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.