Scholarlink

<p>Biopsy images from the colon are necessary to discover cancer, since they display particular layouts of biological elements. Histopathologists use a microscope to inspect the appearance of cells since diagnosing cell deformation is hard and opinion-based. Since opinions differ among human viewers, a sample may receive different cancer grades from one pathologist to another or even from the same person. Consequently, more people are interested in making automated systems for grading images from colon biopsies. Dependable identification at 4X, 10X, and 40X by this approach fostered various exciting developments in computerized examination of colon cells. The authors also highlight a segmentation process that works at any magnification by joining Mask R-CNN with the HED network. Moreover, the system relies on contrast enhancement and color normalization to ensure all cells look the same, making it easier to use features to classify them. Next, hybrid features are created using texture analysis from DT-CWT, deep features from fully connected convolutional networks, and information from morphology and MSER. A set of HOG characteristics is included in the strong features used, providing a more detailed description for the exact identification of cells. Improved Snake Optimization is applied to select the significant features, which boosts the computer’s performance. To summarize, the system takes advantage of the dual-classifier idea by using both RF and LSTM to identify uncommon and repeated data patterns. The proposed fusion of features was found to perform consistently well on the different datasets. Using the AMC and Imediatreat datasets, the model obtained a very high accuracy score and its F1-score reached 98.70%. Despite not achieving high precision, the model performed well on the IPC dataset. Furthermore, the findings from GlaS prove that the model works well, with a precision of 95.00% and an accuracy of 95.50%. Overall, the outcomes point to the feasibility of the model on other data with its high rate of detection and diagnoses.</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

Multi-magnification mask R-CNN and HED-based segmentation with hybrid DTCWT-MSER feature extraction and optimized dual-classifiers for automated colon biopsy malignancy grading

  • 期刊论文

  • 发表日期：

    2025年10月16日

查看全文
  • Valluri Sreelakshmi,
  • D. Vishnu Vardhan

摘要

Biopsy images from the colon are necessary to discover cancer, since they display particular layouts of biological elements. Histopathologists use a microscope to inspect the appearance of cells since diagnosing cell deformation is hard and opinion-based. Since opinions differ among human viewers, a sample may receive different cancer grades from one pathologist to another or even from the same person. Consequently, more people are interested in making automated systems for grading images from colon biopsies. Dependable identification at 4X, 10X, and 40X by this approach fostered various exciting developments in computerized examination of colon cells. The authors also highlight a segmentation process that works at any magnification by joining Mask R-CNN with the HED network. Moreover, the system relies on contrast enhancement and color normalization to ensure all cells look the same, making it easier to use features to classify them. Next, hybrid features are created using texture analysis from DT-CWT, deep features from fully connected convolutional networks, and information from morphology and MSER. A set of HOG characteristics is included in the strong features used, providing a more detailed description for the exact identification of cells. Improved Snake Optimization is applied to select the significant features, which boosts the computer’s performance. To summarize, the system takes advantage of the dual-classifier idea by using both RF and LSTM to identify uncommon and repeated data patterns. The proposed fusion of features was found to perform consistently well on the different datasets. Using the AMC and Imediatreat datasets, the model obtained a very high accuracy score and its F1-score reached 98.70%. Despite not achieving high precision, the model performed well on the IPC dataset. Furthermore, the findings from GlaS prove that the model works well, with a precision of 95.00% and an accuracy of 95.50%. Overall, the outcomes point to the feasibility of the model on other data with its high rate of detection and diagnoses.