Scholarlink

Radiation given to breast cancer patients creates adverse pulmonary effects which can be life threatening. Predicting these adverse effects in the lungs with machine learning algorithms can help improve the quality of life of breast cancer patients. Classification of lung diseases is done using three different Convolutional Neural Network (CNN) architectures with PET-CT fusion images of the lungs. Since the dataset is scarce, pretrained models with Image Net are taken for analysis and transfer learning techniques are applied. ResNet, InceptionV3 and DenseNet121 architectures are tested. Finetuning is applied layer wise in the pretrained model. Results are evaluated with validation accuracy, test accuracy, F1-Score, Precision and Recall. Finetuning of the initial layers provides good results in all the three architectures considered. However, DenseNet121 outperforms other architectures. DenseNet is robust to provide high accuracy scores.

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

High Performance Optimal Transfer Learning Architecture for Classification of Radiation Induced Lung Injury in Breast Cancer Patients

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • S. Praveenkumar,
  • V. Nitin Shreyes,
  • K. B. Jayanthi,
  • C. Rajasekaran,
  • R. Premalatha,
  • R. Sureshkumar

摘要

Radiation given to breast cancer patients creates adverse pulmonary effects which can be life threatening. Predicting these adverse effects in the lungs with machine learning algorithms can help improve the quality of life of breast cancer patients. Classification of lung diseases is done using three different Convolutional Neural Network (CNN) architectures with PET-CT fusion images of the lungs. Since the dataset is scarce, pretrained models with Image Net are taken for analysis and transfer learning techniques are applied. ResNet, InceptionV3 and DenseNet121 architectures are tested. Finetuning is applied layer wise in the pretrained model. Results are evaluated with validation accuracy, test accuracy, F1-Score, Precision and Recall. Finetuning of the initial layers provides good results in all the three architectures considered. However, DenseNet121 outperforms other architectures. DenseNet is robust to provide high accuracy scores.