Scholarlink

<p>Skin cancer is among the most rapidly spreading types of cancer affecting humans. Traditional diagnosis relies on expert dermatologists and specialized equipment, which can be both time-consuming and costly. To reduce diagnostic expenses, deep learning has offered state-of-the-art solutions for early and accurate detection of skin cancer. Nevertheless, analyzing skin lesion images is challenging due to lighting, color, and shape variations. These difficulties make reliable automated recognition of skin cancer essential for enhancing the efficiency of early diagnosis. In response to these challenges, this research introduces a novel approach called customized batch normalization-assisted parallel convolutional neural network for skin cancer classification (CBNPC-SCC), specifically designed to overcome these obstacles and improve diagnostic performance. The proposed approach encompasses four primary stages such as preprocessing, segmentation, feature extraction, and classification phases. Initially, the new pixel estimation-based Wiener filter (NPE-WF) approach is employed for preprocessing the input image, in which the modifications are carried out in the mean, variance, and new pixel computations. This improved version minimizes the noise while enhancing the quality of the image. Subsequently, a channel attention module updated SegNet (CAMSgN) is proposed using the CAM method and is employed for segmenting the preprocessed image, which enhances the capability of isolating the region of interest, enhancing the segmentation of cancer regions. Then, the appropriate features such as local Gabor transitional pattern (LGTrP), multi-texton (MTH) features, threshold-based smoother function in local monotonic pattern (TS-LMP), and deep features are extracted from the segmented outcome. These features effectively capture the transition patterns between pixels and the fine-grained patterns, crucial for differentiating the types of skin lesions. Finally, the obtained features are given to the classification phase, where a hybrid combination of LinkNet and weighted quadratic mean-customized batch normalization-parallel convolutional neural network (WQM-CBN-PCNN) models is proposed for classification. Averaging the outcomes of both classifiers shows the final classification outcomes. Furthermore, the proposed model achieved a higher accuracy of 0.975, <i>F</i>-measure of 0.914, and specificity of 0.986, which surpasses the results of the traditional methods. Overall, the proposed model presents an effective, reliable, and cost-efficient solution for automated skin cancer classification.</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

Skin Cancer Classification Using PCNN-LinkNet Model Trained with Monotonic and Deep Feature Set

  • 期刊论文

  • 发表日期：

    2025年7月29日

查看全文
  • Y. B. Vinay Kumar,
  • H. S. Vimala

摘要

Skin cancer is among the most rapidly spreading types of cancer affecting humans. Traditional diagnosis relies on expert dermatologists and specialized equipment, which can be both time-consuming and costly. To reduce diagnostic expenses, deep learning has offered state-of-the-art solutions for early and accurate detection of skin cancer. Nevertheless, analyzing skin lesion images is challenging due to lighting, color, and shape variations. These difficulties make reliable automated recognition of skin cancer essential for enhancing the efficiency of early diagnosis. In response to these challenges, this research introduces a novel approach called customized batch normalization-assisted parallel convolutional neural network for skin cancer classification (CBNPC-SCC), specifically designed to overcome these obstacles and improve diagnostic performance. The proposed approach encompasses four primary stages such as preprocessing, segmentation, feature extraction, and classification phases. Initially, the new pixel estimation-based Wiener filter (NPE-WF) approach is employed for preprocessing the input image, in which the modifications are carried out in the mean, variance, and new pixel computations. This improved version minimizes the noise while enhancing the quality of the image. Subsequently, a channel attention module updated SegNet (CAMSgN) is proposed using the CAM method and is employed for segmenting the preprocessed image, which enhances the capability of isolating the region of interest, enhancing the segmentation of cancer regions. Then, the appropriate features such as local Gabor transitional pattern (LGTrP), multi-texton (MTH) features, threshold-based smoother function in local monotonic pattern (TS-LMP), and deep features are extracted from the segmented outcome. These features effectively capture the transition patterns between pixels and the fine-grained patterns, crucial for differentiating the types of skin lesions. Finally, the obtained features are given to the classification phase, where a hybrid combination of LinkNet and weighted quadratic mean-customized batch normalization-parallel convolutional neural network (WQM-CBN-PCNN) models is proposed for classification. Averaging the outcomes of both classifiers shows the final classification outcomes. Furthermore, the proposed model achieved a higher accuracy of 0.975, F-measure of 0.914, and specificity of 0.986, which surpasses the results of the traditional methods. Overall, the proposed model presents an effective, reliable, and cost-efficient solution for automated skin cancer classification.