Scholarlink

<p>The purpose of this study provides hybrid approach Concatenated Deep Transfer Learning with Hybridized Optimization Algorithm for early diagnosis and prognosis of oral cancer. Histopathology images are pre-processed using Wiener Filtering to minimize background noise and segmented the images using the U-Net framework. Level features are extracted utilizing discrete wavelet transformation and edge histogram descriptor approaches to enhance image representation. A novel hybrid feature selection approach uses Lasso-Cox model and Fireworks Algorithm (FWA) to optimize accuracy and interpretation during diagnosis. The concatenated deep transfer model utilized NASNet and ShuffleNet architectures to extract features from the same image and use the best features from either model. Following the concatenation of features a fully connected classifier is proposed to store features for greater oral cancer classification accuracy. We also proposed a novel Wild Horse with an integrated Sunflower Optimization Algorithm (WH-SOA) ability to optimize model hyperparameters for oral squamous cell carcinoma classification. We validated our experiments in MATLAB with results for model that achieved a precision score of 0.84786 indicating strong identification of positive cases. Overall, this integrated approach enhanced the opportunity for timely detection of oral cancer cases that enhance clinical decision support and ultimately improve patient health outcomes.</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

Concatenated Deep Transfer Learning with Hybridized Optimization Algorithm for Early Prediction and Diagnosis of Oral Cancer Disease

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Alka Kumari,
  • Ashok Kumar Mehta,
  • Danish Ali Khan

摘要

The purpose of this study provides hybrid approach Concatenated Deep Transfer Learning with Hybridized Optimization Algorithm for early diagnosis and prognosis of oral cancer. Histopathology images are pre-processed using Wiener Filtering to minimize background noise and segmented the images using the U-Net framework. Level features are extracted utilizing discrete wavelet transformation and edge histogram descriptor approaches to enhance image representation. A novel hybrid feature selection approach uses Lasso-Cox model and Fireworks Algorithm (FWA) to optimize accuracy and interpretation during diagnosis. The concatenated deep transfer model utilized NASNet and ShuffleNet architectures to extract features from the same image and use the best features from either model. Following the concatenation of features a fully connected classifier is proposed to store features for greater oral cancer classification accuracy. We also proposed a novel Wild Horse with an integrated Sunflower Optimization Algorithm (WH-SOA) ability to optimize model hyperparameters for oral squamous cell carcinoma classification. We validated our experiments in MATLAB with results for model that achieved a precision score of 0.84786 indicating strong identification of positive cases. Overall, this integrated approach enhanced the opportunity for timely detection of oral cancer cases that enhance clinical decision support and ultimately improve patient health outcomes.