Scholarlink

<p>Deep learning has demonstrated enormous promise in processing of medical images. A major source of cancer-related mortality globally is colon cancer, demands efficient and accurate diagnostic tools. This study introduces a lightweight, high-performance deep learning model based on MobileNet architecture to classify colon tissue images as either adenocarcinoma (cancerous) or normal. The model leverages pre-trained ImageNet weights for feature extraction and employs data augmentation techniques, like rescaling, flipping, zooming, and shearing, to improve generalization. Trained on 800 high-resolution images with equal representation of both classes, model attained a training accuracy ranging from 93.25% to 99.87% and a validation accuracy of up to 96.88%. These results demonstrate the model’s robustness and suitability for scenarios with limited resources, making it a viable candidate in real-time clinical applications. By addressing challenges such as computational efficiency and variability in medical images, this work contributes significantly to advancing automated colon cancer diagnostics and improving patient care.</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

Robust Deep Learning Approach for Colon Cancer Detection Using MobileNet

  • 期刊论文

  • 发表日期：

    2025年3月28日

查看全文
  • Tushar H. Jaware,
  • Jitendra P. Patil,
  • Ravindra D. Badgujar

摘要

Deep learning has demonstrated enormous promise in processing of medical images. A major source of cancer-related mortality globally is colon cancer, demands efficient and accurate diagnostic tools. This study introduces a lightweight, high-performance deep learning model based on MobileNet architecture to classify colon tissue images as either adenocarcinoma (cancerous) or normal. The model leverages pre-trained ImageNet weights for feature extraction and employs data augmentation techniques, like rescaling, flipping, zooming, and shearing, to improve generalization. Trained on 800 high-resolution images with equal representation of both classes, model attained a training accuracy ranging from 93.25% to 99.87% and a validation accuracy of up to 96.88%. These results demonstrate the model’s robustness and suitability for scenarios with limited resources, making it a viable candidate in real-time clinical applications. By addressing challenges such as computational efficiency and variability in medical images, this work contributes significantly to advancing automated colon cancer diagnostics and improving patient care.