Multimodal Neural Network Model for Intelligent Diagnosis of Melanoma Skin Disease

A melanoma skin disease cure is more likely with early identification. Melanoma skin diagnosis by dermoscopy-based manual visual inspection has $75 \%$ to $\mathbf{8 0 \%}$ accuracy but low efficiency. Doctor skill and experience are crucial to this treatment. Image input and metadata are used to create an Intelligent Model for multimodal neural networks in the study. Metadata includes perceptron learning model feature vectors with patient information, lesion collection locations, image resolution, and quantity. Image data is CNN-derived feature vectors. Disease categorization by fusing and mapping these two feature vectors allows early auxiliary melanoma diagnosis. 58,457 2019 and 2020 ISIC sample records were compiled and organized. Training and testing data were split 4:1, and experiments compared the convolutional neural network technique to the multimodal Model. Based on the results, the suggested strategy may improve melanoma auxiliary diagnosis classification models by $1 \%$.