Melanoma Detection Using Egret Search Golden Optimization Tuned Distributed Pooling-Based Fused Deep BiLSTM Model

Melanoma, the most lethal type of skin cancer, remains a formidable obstacle in the quest for early detection. Despite significant strides in the field, numerous enduring challenges obstruct the effectiveness of current models and techniques. A critical disadvantage of current melanoma detection models is their limited sensitivity, particularly in identifying early-stage melanomas. However, there exists the complexity in visual features of skin lesion images due to the nonhomogenous features and fuzzy boundaries that limit their performance. Additionally, the reliability of melanoma detection is degraded by the noise, shadows, and artifacts present in the medical images. This research addresses these challenges and introduces a novel melanoma detection model that ultimately enhances diagnostic accuracy and enables timely intervention. This research aims to develop an Egret Search Golden (ESG) optimization tuned distributed pooling-based BiLSTM model (ESG-distributed pooling-based BiLSTM) model to enhance the detection accuracy of melanoma through an innovative approach. It begins by gathering data and then employs a series of advanced techniques. The adaptive optimized Otsu thresholding process provides more precise, segmenting of the melanoma lesions accurately from the background, and feature extraction encompassing texture patterns, deep features, and hybrid structural features that provide effective melanoma detection. The extracted features utilizing the Gray Level Co-occurrence Matrix are then fed into a sophisticated distributed pooling-based fused deep BiLSTM model, further fine-tuned using the ESG approach, to overcome all the challenges of the early detection of melanoma, a unique optimization method provides high convergence rate, effective handling of high-dimensional problems, robustness, and reduced the time complexity. To extract the features, the encoder and decoder subnetworks are associated in a sequence pathway to bring the semantic features and provide efficiency in melanoma detection in the initial stage. Specifically, for the Melanoma Skin Cancer Dataset, the model yielded impressive TP 90 values, showcasing metrics at 96.13%, 96.06%, and 96.10%, respectively. Moreover, when adopting the [Formula: see text]-fold 6 approach, the model’s performance elevated to even higher levels, delivering better measures of 96.40%, 96.27%, and 96.05% for accuracy, sensitivity, and specificity.