Scholarlink

<p>Life-critical and patient-safety-related decision-making is very significant in machine intelligence for image recognition with rigorous certainty, as visual elements are represented by pixels. High-accuracy feature extraction is a key factor for certainty prediction in machine intelligence systems. Convolutional neural networks (CNNs) are the widespread method for learning high-level characteristics. However, limited feature information and incorrect consideration weighting impede the certainty of CNNs. Most CNNs do not consider low-level and high-level information, leading to prediction uncertainty. Feature information in hidden layers provides a great opportunity for feature discrimination. Researchers have not paid enough attention to the capabilities of hidden layers in CNN, and instead, they have concentrated on the last feature map of the final convolution layer. The image dimension problem is a big issue for uncertainty prediction, and combining high-level and low-level features is essential to addressing this issue. The primary focus of this research is on how to utilize the feature information of hidden layers in the CNN framework to eliminate the uncertainty of the CNN algorithm. This research built a predictive uncertainty model by concatenating features from different hidden layers in CNN and other advanced CNNs. The Bayesian approach is used to quantify uncertainty and measure predictive uncertainty. This proposed hybrid CNN model, which is based on the fusion of multi-feature maps and the Bayesian approach, minimizes the uncertainty while improving performance better than VGG16 and DenseNet121. This novel model achieves 97.05% accuracy and 2.22% predictive uncertainty. This unique model predicts COVID-19 in 99.13%, normal in 94.32%, and pneumonia in 97.78% based on the findings of the F1-score, confusion matrix, and ROC-AUC (Receiver Operating Characteristic curve and its Area Under the Curve). Furthermore, this model demonstrates that the third, fourth, fifth, sixth, and seventh layers are critical for confidence prediction. Mining hidden layers in CNN for certainty prediction in medical imaging.</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

Mining hidden layers in CNN for certainty prediction in medical imaging

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Y. M. Hirimutugoda,
  • T. P. Sliva,
  • N. M. Wagarachchi

摘要

Life-critical and patient-safety-related decision-making is very significant in machine intelligence for image recognition with rigorous certainty, as visual elements are represented by pixels. High-accuracy feature extraction is a key factor for certainty prediction in machine intelligence systems. Convolutional neural networks (CNNs) are the widespread method for learning high-level characteristics. However, limited feature information and incorrect consideration weighting impede the certainty of CNNs. Most CNNs do not consider low-level and high-level information, leading to prediction uncertainty. Feature information in hidden layers provides a great opportunity for feature discrimination. Researchers have not paid enough attention to the capabilities of hidden layers in CNN, and instead, they have concentrated on the last feature map of the final convolution layer. The image dimension problem is a big issue for uncertainty prediction, and combining high-level and low-level features is essential to addressing this issue. The primary focus of this research is on how to utilize the feature information of hidden layers in the CNN framework to eliminate the uncertainty of the CNN algorithm. This research built a predictive uncertainty model by concatenating features from different hidden layers in CNN and other advanced CNNs. The Bayesian approach is used to quantify uncertainty and measure predictive uncertainty. This proposed hybrid CNN model, which is based on the fusion of multi-feature maps and the Bayesian approach, minimizes the uncertainty while improving performance better than VGG16 and DenseNet121. This novel model achieves 97.05% accuracy and 2.22% predictive uncertainty. This unique model predicts COVID-19 in 99.13%, normal in 94.32%, and pneumonia in 97.78% based on the findings of the F1-score, confusion matrix, and ROC-AUC (Receiver Operating Characteristic curve and its Area Under the Curve). Furthermore, this model demonstrates that the third, fourth, fifth, sixth, and seventh layers are critical for confidence prediction. Mining hidden layers in CNN for certainty prediction in medical imaging.