An Intuitive U-LSTM Model for Classification and Recognition for Skin Cancer Detection

The precise identification and categorization of cutaneous lesions and melanoma are of paramount importance in the timely discovery and successful management of these conditions. This paper presents a unique methodology that integrates the Intuitive U-Net and Long Short-Term Memory (LSTM) architecture to achieve precise categorization of skin lesions and melanoma. The Intuitive U-Net architecture, which draws inspiration from the U-Net model, has been specifically developed to effectively capture complex characteristics and complicated patterns present in dermoscopic images. Skip connections are employed in order to maintain spatial information and retrieve pertinent features at various scales. This improves the model’s capacity to discern distinct categories of lesions and effectively categorize instances of melanoma. In order to enhance the precision of temporal relationships among sequences of images, the network is augmented with LSTM units. The Long Short-Term Memory (LSTM) architecture facilitates the model in taking into account the sequential context of images, thereby capturing temporal changes and patterns. Monitoring the advancement of skin lesions and the evolution of melanoma holds significant importance in this context. The methodology we propose is assessed using a heterogeneous dataset that includes a range of skin lesions and instances of melanoma. The proposed findings illustrate the efficacy of the Intuitive U-Net and LSTM architecture in accurately categorizing skin lesions and melanoma using transforming modelling with the U-LSTM model. The integration of feature extraction from pictures with temporal modelling using Long Short-Term Memory (LSTM) has been found to enhance sensitivity, specificity, and accuracy in comparison to current methodologies. With this approach, we demonstrate the overall design model with different specification parameters indicating the effectiveness of classification and identification with an accuracy of 91.5% on lesions while melanoma with 98.2%. The other performance metrics, such as sensitivity, specificity, and ROC curves, are utilized to emphasize providing the best classification model feature performance when compared to state-of-the-art architectures.