Scholarlink

<p>Bone cancer is a cancerous tumour that originates in bone tissue and causes inflammation, pain, and fractures. It could spread from other body areas or start in the bones. After a diagnosis, it is essential to begin appropriate treatment procedures at once, especially for malignant bone tumours. Several techniques were developed for bone cancer detection at an earlier stage. However, those techniques failed to provide efficient results due to their overfitting, higher false positive rate, etc. To overcome these kinds of issues, the proposed methodology is introduced to provide efficient performance. Initially, the data were gathered from Chonnam National University Hospital (CNUH) dataset. The pre-processing was done using the uprated non-local mean filtering (Up-NonMF) and contrast-limited ADAPTIVE histogram equalisation (CLAHE) for noise reduction and contrast enhancement. Then, the multi-stage UNet (Multi-SUnet) is used for segmentation. Finally, the proposed self-attention-based dense cascaded capsule network (SADCCNet) model is used to classify bone cancer. Here, the SADCCNet model is optimised using the binary zebra optimisation algorithm (BZOA) to improve its efficiency. This model is executed using the Python tool. The performance of the proposed technique is determined by comparing it with various existing models such as Res_UNet, MobiNet_V2, and VGG-16 models. The proposed model achieves 98.93% accuracy, 97.05% precision, 97.09% recall, 97% F1 score, and 97.35% Specificity.</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

SADCCNet: self-attention-based dense cascaded capsule network for bone cancer detection using deep learning approach

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • Jameer Gulab Kotwal,
  • Nand Kishore Sharma,
  • Rachna Somkunwar,
  • Dnyaneshwar Bavkar,
  • Swapnil Deshmukh,
  • Vajid Khan,
  • Nilesh Korade

摘要

Bone cancer is a cancerous tumour that originates in bone tissue and causes inflammation, pain, and fractures. It could spread from other body areas or start in the bones. After a diagnosis, it is essential to begin appropriate treatment procedures at once, especially for malignant bone tumours. Several techniques were developed for bone cancer detection at an earlier stage. However, those techniques failed to provide efficient results due to their overfitting, higher false positive rate, etc. To overcome these kinds of issues, the proposed methodology is introduced to provide efficient performance. Initially, the data were gathered from Chonnam National University Hospital (CNUH) dataset. The pre-processing was done using the uprated non-local mean filtering (Up-NonMF) and contrast-limited ADAPTIVE histogram equalisation (CLAHE) for noise reduction and contrast enhancement. Then, the multi-stage UNet (Multi-SUnet) is used for segmentation. Finally, the proposed self-attention-based dense cascaded capsule network (SADCCNet) model is used to classify bone cancer. Here, the SADCCNet model is optimised using the binary zebra optimisation algorithm (BZOA) to improve its efficiency. This model is executed using the Python tool. The performance of the proposed technique is determined by comparing it with various existing models such as Res_UNet, MobiNet_V2, and VGG-16 models. The proposed model achieves 98.93% accuracy, 97.05% precision, 97.09% recall, 97% F1 score, and 97.35% Specificity.