ABSTRACT
Background: Skin cancer is the predominant kind of cancer, with initial detection being vital for effective treatment. Traditional methods of detection often rely on visual inspection by dermatologists, which can be time-consuming and subjective. Recent advancements in deep learning and transfer learning offer promising avenues for enhancing precision and efficacy of skin cancer finding.
Aim: This study aimed to enhance skin cancer recognition by leveraging multi-scale deep learning and transfer learning techniques.
Methods: A dataset comprising images of skin lesions was collected from 90 patients over a study duration spanning from March 2023 to February 2024. Multi-scale deep learning architectures were trained on this dataset to extract hierarchical features from different resolutions of the input images. Transfer learning procedures were employed to fine-tune pre-trained models on skin lesion dataset, enabling the models to generalize well even with limited training data.
Results: The multi-scale deep learning models, combined with transfer learning, demonstrated superior performance in detecting skin cancer lesions compared to traditional methods. The models achieved a high level of accuracy and robustness across various types of skin lesions and patient demographics.
Conclusion: This study highlights the efficacy of employing multi-scale deep learning and transfer learning techniques for enhancing skin cancer detection. The integration of these advanced methods has possibility to revolutionize the early diagnosis and treatment of skin cancer, ultimately improving patient outcomes.
Keywords: Skin cancer detection, Multi-scale deep learning, Transfer learning, Image classification, Early diagnosis.