Scholarlink

<p>Colorectal cancer (CRC) segmentation is a difficult task because of the structural complexity of polyps in colonoscopy and endoscopy imaging. This paper proposes CRCNet, a novel deep learning (DL) framework to address these challenges using specifically designed components that improve feature learning and localization precision. CRCNet combines a convolutional multi-layer perceptron (MLP) encoder that preserves both spatial and contextual representations; a synchronized feature boost (SFB) module to maintain fine-grained low-level features usually lost during downsampling. A convolutional block attention module (CBAM) to focus on diagnostically relevant areas using spatial and channel-wise attention. Further, a semantic-local feature aggregation (SLFA) block merges semantic context with localized information to enhance boundary clarity in the decoder. CRCNet is evaluated on four benchmark datasets: CVC-ClinicDB, ETIS, CVC-ColonDB, and CVC-300. It achieves dice coefficient of 0.9519, 0.9139, 0.8677, and 0.9421 intersection over union scores of 0.9083, 0.8438, 0.7692, and 0.8909, respectively. The model also demonstrates exceptional boundary accuracy and lowers mean absolute error (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12530_2025_9716_Article_IEq1.gif" Format="GIF" Height="15" Rendition="HTML" Resolution="72" Type="Linedraw" Width="65" /> </InlineMediaObject> <EquationSource Format="TEX">\(\le 0.0180\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>≤</mo> <mn>0.0180</mn> </mrow> </math></EquationSource> </InlineEquation>) with a structure measure (Sm) above 0.93 and an enhanced measure above 0.95 across all datasets. The lightweight architecture (1.18&#xa0;M parameters) of CRCNet ensures computational efficiency by preserving detailed anatomy details in low-contrast areas. These experimental results show the superior performance of CRCNet over existing techniques by focusing on its adaptability in managing various polyp shapes, scale modifications, and imaging artefacts. CRCNet improves the accuracy of CRC diagnosis by fusing attention systems and advanced feature enhancement. </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

CRCNet: convolutional multi-layer perceptron encoder with attention module for colorectal cancer segmentation

  • 期刊论文

  • 发表日期：

    2025年7月5日

查看全文
  • Manoj Kumar Singh,
  • Satish Chand,
  • Devender Kumar

摘要

Colorectal cancer (CRC) segmentation is a difficult task because of the structural complexity of polyps in colonoscopy and endoscopy imaging. This paper proposes CRCNet, a novel deep learning (DL) framework to address these challenges using specifically designed components that improve feature learning and localization precision. CRCNet combines a convolutional multi-layer perceptron (MLP) encoder that preserves both spatial and contextual representations; a synchronized feature boost (SFB) module to maintain fine-grained low-level features usually lost during downsampling. A convolutional block attention module (CBAM) to focus on diagnostically relevant areas using spatial and channel-wise attention. Further, a semantic-local feature aggregation (SLFA) block merges semantic context with localized information to enhance boundary clarity in the decoder. CRCNet is evaluated on four benchmark datasets: CVC-ClinicDB, ETIS, CVC-ColonDB, and CVC-300. It achieves dice coefficient of 0.9519, 0.9139, 0.8677, and 0.9421 intersection over union scores of 0.9083, 0.8438, 0.7692, and 0.8909, respectively. The model also demonstrates exceptional boundary accuracy and lowers mean absolute error ( \(\le 0.0180\) ≤ 0.0180 ) with a structure measure (Sm) above 0.93 and an enhanced measure above 0.95 across all datasets. The lightweight architecture (1.18 M parameters) of CRCNet ensures computational efficiency by preserving detailed anatomy details in low-contrast areas. These experimental results show the superior performance of CRCNet over existing techniques by focusing on its adaptability in managing various polyp shapes, scale modifications, and imaging artefacts. CRCNet improves the accuracy of CRC diagnosis by fusing attention systems and advanced feature enhancement.