Scholarlink

<p>Grading of Invasive Ductal Carcinoma (IDC) by trained pathologists is a complex and subjective task, often leading to significant inter-observer variability. Accurate and consistent grading is crucial as it directly influences the patient’s treatment plan and survival outcomes. Existing automated techniques, however, often suffer from limitations such as suboptimal accuracy and performance metrics and a lack of robustness across diverse datasets. To address these challenges, this study introduces a novel Automated Grading System (AGS) utilizing the Rank Expansion Network (RexNet) deep learning architecture. The proposed system incorporates innovative techniques at three critical design stages—model architecture, training methodology, and inference optimization—to enhance performance. Specifically, the Gradient Accumulation algorithm enables efficient training on modestly resourced machines, reducing computational barriers. The system achieves state-of-the-art performance with an accuracy of <b>96.41%</b>, AUC of <b>0.99</b>, and F1-score of <b>95</b>% on the Databiox dataset, significantly surpassing existing methods. Additionally, it demonstrates robustness and adaptability by achieving <b>96.84%</b> <b>accuracy</b> on the Agios Pavlos dataset. This work highlights a step forward in developing reliable and scalable tools for IDC grading, with potential for clinical application.</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

An end-to-end deep rank expansion network for automated grading of breast carcinoma using histopathology images with gradient accumulation

  • 期刊论文

  • 发表日期：

    2025年3月18日

查看全文
  • Vivek Harshey,
  • Amar Partap Singh Pharwaha

摘要

Grading of Invasive Ductal Carcinoma (IDC) by trained pathologists is a complex and subjective task, often leading to significant inter-observer variability. Accurate and consistent grading is crucial as it directly influences the patient’s treatment plan and survival outcomes. Existing automated techniques, however, often suffer from limitations such as suboptimal accuracy and performance metrics and a lack of robustness across diverse datasets. To address these challenges, this study introduces a novel Automated Grading System (AGS) utilizing the Rank Expansion Network (RexNet) deep learning architecture. The proposed system incorporates innovative techniques at three critical design stages—model architecture, training methodology, and inference optimization—to enhance performance. Specifically, the Gradient Accumulation algorithm enables efficient training on modestly resourced machines, reducing computational barriers. The system achieves state-of-the-art performance with an accuracy of 96.41%, AUC of 0.99, and F1-score of 95% on the Databiox dataset, significantly surpassing existing methods. Additionally, it demonstrates robustness and adaptability by achieving 96.84% accuracy on the Agios Pavlos dataset. This work highlights a step forward in developing reliable and scalable tools for IDC grading, with potential for clinical application.