Scholarlink

<p>Skin cancers, particularly melanoma are significant health issues. Early identification of skin disease models is implemented with various dermoscopic examinations, medical screening, and so on. Yet, this traditional model requires assistance from medical experts to offer more accurate diagnosis outcomes. Moreover, identifying skin disease at the starting stages is crucial for enhancing the treatment plans of the victims. Recently, computerized techniques known as computer-aided diagnosis applications have been employed to perform early prognosis of skin diseases. Additionally, artificial intelligence is utilized to carry out automatic detection of skin diseases via skin images. Still, automatic diagnosis of skin cancers at the beginning stages is hard because of the lack of contrast between the moles of melanoma and skin areas as well as the high degree of color similarity among the affected and unaffected regions within the skin. Currently, deep learning techniques provide promising outcomes in medical imaging. Thus, the research work implements a novel skin cancer detection model is presented using deep learning. The standard datasets are considered for data collection process. The total image acquired from both dataset 1 and dataset 2 contains 5000 and 200 images. In dataset 1, the total 5000 image are divided into 3750 for training and 1250 for testing. Likewise, a total 200 images of dataset 2 are divided into 150 images for training and 50 images for testing. Then, the collected skin image is fed to the abnormality segmentation process. Here, the Yolo-Unet++ with LovaszSoftMax with IoU Loss Function (Y-UNet++-LS-IoULF) is utilized for segmenting the images. Later, the segmented images are forwarded to the vision transformer (ViT)-aided Adaptive EfficientB7 network (ViT-AEB7) for detecting skin cancer. Here, an updated random variable-aided Kookaburra optimization algorithm (URV-KOA) is used for tuning the network parameters of the EfficientB7 model. Lastly, an empirical investigation is executed for the suggested system by comparing it with classical-related approaches. The findings of the recommended method attain 89.91% and 84.92% in terms of F1-score and MCC measure. Thus, the significant advancement in the developed model attains better detected outcomes and enables accurate treatment planning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel framework of skin cancer detection using Yolo-Unet++ segmentation model with adaptive deep learning-based classification

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • S. Jenita Christy,
  • G. Rosline Nesa Kumari

摘要

Skin cancers, particularly melanoma are significant health issues. Early identification of skin disease models is implemented with various dermoscopic examinations, medical screening, and so on. Yet, this traditional model requires assistance from medical experts to offer more accurate diagnosis outcomes. Moreover, identifying skin disease at the starting stages is crucial for enhancing the treatment plans of the victims. Recently, computerized techniques known as computer-aided diagnosis applications have been employed to perform early prognosis of skin diseases. Additionally, artificial intelligence is utilized to carry out automatic detection of skin diseases via skin images. Still, automatic diagnosis of skin cancers at the beginning stages is hard because of the lack of contrast between the moles of melanoma and skin areas as well as the high degree of color similarity among the affected and unaffected regions within the skin. Currently, deep learning techniques provide promising outcomes in medical imaging. Thus, the research work implements a novel skin cancer detection model is presented using deep learning. The standard datasets are considered for data collection process. The total image acquired from both dataset 1 and dataset 2 contains 5000 and 200 images. In dataset 1, the total 5000 image are divided into 3750 for training and 1250 for testing. Likewise, a total 200 images of dataset 2 are divided into 150 images for training and 50 images for testing. Then, the collected skin image is fed to the abnormality segmentation process. Here, the Yolo-Unet++ with LovaszSoftMax with IoU Loss Function (Y-UNet++-LS-IoULF) is utilized for segmenting the images. Later, the segmented images are forwarded to the vision transformer (ViT)-aided Adaptive EfficientB7 network (ViT-AEB7) for detecting skin cancer. Here, an updated random variable-aided Kookaburra optimization algorithm (URV-KOA) is used for tuning the network parameters of the EfficientB7 model. Lastly, an empirical investigation is executed for the suggested system by comparing it with classical-related approaches. The findings of the recommended method attain 89.91% and 84.92% in terms of F1-score and MCC measure. Thus, the significant advancement in the developed model attains better detected outcomes and enables accurate treatment planning.