Scholarlink

<p>The dangerous skin disease known as Melanoma ranks as one of the most lethal globally when it comes to cancer types. The likelihood of curing the disease increases significantly through early diagnosis. The presence of visual similarities between melanoma and benign cases makes it difficult to distinguish them in dermoscopy images. The proposed evaluation system integrates a complex deep-learning framework that functions as a Computer-Aided Diagnosis (CAD) system. The method is divided into boundary tracing and type determination of skin irregularities as its two main processes. The skin lesion boundaries in dermoscopy pictures get traced at the first step by implementing a CNN-based DoubleU-Net model. After segmentation, the model utilizes advanced convolutional classifiers, including EfficientNetB7, DenseNet201, InceptionV3, ResNet152V2 and InceptionResNetV2, in order to categorize the lesions. Segmentation of the lesions enables specific features in the images to be retrieved. These classifiers are optimised by integrating ISIC 2018 and ISIC 2019 datasets, which present balanced combined segments of augmented information. The evaluation of performance indicators, which included AUC score and precision and recall, enabled an assessment between these classifiers. Tests proved that classifiers perform better when operating on separated data streams. The classifiers demonstrated AUC performance at 92.39% and 93.91%, 92.08%, 90.92%, and 94.98% when evaluated. When evaluated, the InceptionResnetV2 delivered the best classification results.</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

Deep learning approaches for detecting malignant melanoma in dermoscopic imagery

  • 期刊论文

  • 发表日期：

    2025年7月26日

查看全文
  • Abhishek Thakur,
  • Sudhansu Kumar Mishra,
  • Monica Bhutani,
  • Prajna Parimita Dash,
  • Prabhat Kumar Upadhyay,
  • Sitanshu Sekhar Sahu

摘要

The dangerous skin disease known as Melanoma ranks as one of the most lethal globally when it comes to cancer types. The likelihood of curing the disease increases significantly through early diagnosis. The presence of visual similarities between melanoma and benign cases makes it difficult to distinguish them in dermoscopy images. The proposed evaluation system integrates a complex deep-learning framework that functions as a Computer-Aided Diagnosis (CAD) system. The method is divided into boundary tracing and type determination of skin irregularities as its two main processes. The skin lesion boundaries in dermoscopy pictures get traced at the first step by implementing a CNN-based DoubleU-Net model. After segmentation, the model utilizes advanced convolutional classifiers, including EfficientNetB7, DenseNet201, InceptionV3, ResNet152V2 and InceptionResNetV2, in order to categorize the lesions. Segmentation of the lesions enables specific features in the images to be retrieved. These classifiers are optimised by integrating ISIC 2018 and ISIC 2019 datasets, which present balanced combined segments of augmented information. The evaluation of performance indicators, which included AUC score and precision and recall, enabled an assessment between these classifiers. Tests proved that classifiers perform better when operating on separated data streams. The classifiers demonstrated AUC performance at 92.39% and 93.91%, 92.08%, 90.92%, and 94.98% when evaluated. When evaluated, the InceptionResnetV2 delivered the best classification results.