A Review on Transfer Learning Methods Used for Skin Cancer Detection

Abstract

Skin cancer is the most common type, with early detection and diagnosis crucial for improving patient outcomes. Machine learning algorithms, particularly deep learning approaches such as convolutional neural networks (CNNs), have been shown to achieve high levels of accuracy in image classification tasks and have been applied in skin cancer identification. Transfer learning, a technique that allows a model trained on one task to be fine-tuned for a related task, has also been widely applied in medical image analysis and potentially significantly reduces the amount of labelled data required for training. This review aims to synthesise the current state of knowledge on using transfer learning techniques for skin cancer identification, evaluate their performance, and identify any remaining challenges or limitations in the field. The review found that transfer learning techniques, particularly CNNs, have demonstrated high levels of accuracy in skin cancer identification and that the use of a variety of datasets suggests that these techniques may be able to generalise to different types of skin cancer and patient populations. The most used datasets are the HAM1000 and ISIC datasets, and clinical data is also being used. The results indicate that transfer learning techniques have demonstrated high levels of accuracy in skin cancer identification, and the use of transfer learning has the potential to significantly reduce the amount of labelled data required for training. Challenges and limitations in the field include the lack of consensus on performance measures and the limited availability of labelled data. Further research is needed to address these challenges and evaluate the generalizability and impact of transfer learning models in real-world clinical settings.