Scholarlink

<p>With the development of deep learning, the integration of deep learning techniques to the histopathological description of images has transformed the sphere of diagnostic pathology allowing the analysis of tissue specimens with high accuracy, a huge load and in an automated fashion. The deep learning-based systems have become the main breakthrough in the sphere of diagnostic pathology that allows the automatic evaluation of the human tissue samples at the highest possible level of accuracy. This work provides a new and generalizable transfer learning model that is aimed at dramatically increasing the classification of H&amp;E-stained histological images. We evaluated and compared the state of the art pre-trained architecture DenseNet121, InceptionResNetV2, Xception, MobileNetV2, NASNetMobile,ResNet50V2, and the Vision Transformer (ViT) both in terms of application and optimization of transfer learning strategies. The fine-tuned models obtained were then combined through advanced ensemble methods (weighted averaging and model averaging) to put the maximum weight on predictive stability. More importantly, this transfer learning regimen gave optimum results in all models. The use of a strict k-fold cross-validation methodology ensured the strength and the external validity of the framework using a variety of histopathological specimens. Empirical results of our tests show that the proposed ensemble approach is vastly superior to the performance of individual models with a validation accuracy of 99.1 and F1-score of 0.9908. These findings are a significant advancement compared to current contemporary approaches, which are successful in overcoming the problem of limited data and complicated tissue architecture. This publication develops the fields of computational histopathology leading to more precise and scalable automated diagnostics of diseases and cancer grades, in clinical pathology.</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

Histopathological image classification for enhanced diagnosis and prognosis through transfer learning and ensemble vision models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • Nirjhar Gope,
  • Nazmus Sakib,
  • Mohammad Anwar Hosen

摘要

With the development of deep learning, the integration of deep learning techniques to the histopathological description of images has transformed the sphere of diagnostic pathology allowing the analysis of tissue specimens with high accuracy, a huge load and in an automated fashion. The deep learning-based systems have become the main breakthrough in the sphere of diagnostic pathology that allows the automatic evaluation of the human tissue samples at the highest possible level of accuracy. This work provides a new and generalizable transfer learning model that is aimed at dramatically increasing the classification of H&E-stained histological images. We evaluated and compared the state of the art pre-trained architecture DenseNet121, InceptionResNetV2, Xception, MobileNetV2, NASNetMobile,ResNet50V2, and the Vision Transformer (ViT) both in terms of application and optimization of transfer learning strategies. The fine-tuned models obtained were then combined through advanced ensemble methods (weighted averaging and model averaging) to put the maximum weight on predictive stability. More importantly, this transfer learning regimen gave optimum results in all models. The use of a strict k-fold cross-validation methodology ensured the strength and the external validity of the framework using a variety of histopathological specimens. Empirical results of our tests show that the proposed ensemble approach is vastly superior to the performance of individual models with a validation accuracy of 99.1 and F1-score of 0.9908. These findings are a significant advancement compared to current contemporary approaches, which are successful in overcoming the problem of limited data and complicated tissue architecture. This publication develops the fields of computational histopathology leading to more precise and scalable automated diagnostics of diseases and cancer grades, in clinical pathology.