Enhanced melanoma and non-melanoma skin cancer classification using a hybrid LSTM-CNN model

Melanoma is the most dangerous type of skin cancer. Although it accounts for only about 1% of all skin cancer cases, it is responsible for the majority of skin cancer-related deaths. Early detection and accurate diagnosis are crucial for improving the prognosis and survival rates of patients with melanoma. This paper presents a novel approach for the automatic identification of cutaneous lesions by integrating convolutional neural networks (CNNs) with long short-term memory (LSTM) networks. In the proposed approach, the image of each skin lesion is divided into a sequence of tags of a particular size, which is then treated by the LSTM network to capture temporal dependence and relevant relationships between different spatial regions. This patching sequence allows the modeling system to analyze the local pattern in the image. Time CNN layers are later used to extract spatial functions, such as texture, edges, and color variation, on each patch. A Softmax layer is then used for classification, providing a probability distribution over the possible classes. We use the HAM10000 dataset, which contains 10,015 skin lesion images. Experimental results demonstrate that the proposed method outperforms recent models in several metrics, including accuracy, recall, precision, F1 score, and ROC curve performance.