Classification of Skin Diseases in the Internet of Medical Things using Hybrid Deep Learning
Dandu Madhavi Latha, K. Anusha, Rafath Samrin, Pundru Chandra Shaker Reddy, Bolleddu Sudhakar, Goski Sathish · 2024
When it comes to the IoMT, skin lesion examination is absolutely essential for making an accurate diagnosis. Skin lesion analysis relies heavily on Computer-aided design(CAD) technologies to enhance accuracy and efficiency. Using hybrid deep learning approaches, this work focuses on segmenting and classifying skin lesions from dermoscopy images. Mask Region-based Convolutional Neural Network (MRCNN) for semantic separation and ResNet50 for lesion-detection are two state-of-the-art methodologies that are combined in this research’s hybrid deep learning model. A skin lesion’s exact location can be determined with the use of the MRCNN’s border delineation capabilities. For comprehensive model training, we gather a large collection of dermoscopy photos that have been annotated. This dataset is used to train a hybrid deep-learning(DL) model that can capture images’ nuanced representations. An accuracy rate of 95.49% in segmenting lesions into different groups demonstrates the model’s capability to do so. Another significant improvement over older method is the high reliability and accuracy of skin lesion classification. After being tested on the ISIC2020 Challenge dataset, the model achieved an impressive 96.75% accuracy rate. Classification and segmentation models outperform the state-of-the-art in Internet-of-Medical-Things(IoMT). When it comes to segmenting and classifying skin lesions, this paper’s hybrid deep learning approach works wonders. Outperforming the existing gold standards, the results demonstrate that the model can enhance diagnostic accuracy in an IoMT environment.