Scholarlink

Breast cancer is one of the most common types of cancer, particularly among women, and it leads the list each year when compared to other types of cancer worldwide. The only way to bring the escalating number of cases under control is by detecting the condition at an early stage. It can be detected using numerous imaging modalities, but the images created by each modality are far more complex, causing exhaustion in medical personnel due to the high number of cases, leading to more false positive predictions. Robust deep CNN models are capable of comprehending these complex images. The proposed work presents a novel hybrid ERCNet model that was designed by integrating the finest aspects of EfficientNetv2Small, ResNet50, and Capsule Network. The EfficientNetv2Small and ResNet50 models are concatenated to extract features from the multimodality images, while the capsule network is used to record the spatial hierarchies between features. The model is being trained and tested on multiple imaging modalities, including histopathology, mammography and ultrasound images. The model has performed well, attaining an average accuracy of 99.6% and an average loss of 0.012. Individual outcomes were also good, with histology accuracy of 100%, mammography accuracy of 99.7%, and ultrasound imaging accuracy of 99.1%, all with good precision, recall, and F1-score. All these results suggest that the proposed model ERCNet can not only be used in breast cancer diagnosis but can be applied in any medical imaging task in future.

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

Novel Hybrid Deep Learning-Based ERCNet Framework for Enhanced Breast Cancer Identification Using Multi-Modal Imaging Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Saket Kumar Singh,
  • K. Sridhar Patnaik

摘要

Breast cancer is one of the most common types of cancer, particularly among women, and it leads the list each year when compared to other types of cancer worldwide. The only way to bring the escalating number of cases under control is by detecting the condition at an early stage. It can be detected using numerous imaging modalities, but the images created by each modality are far more complex, causing exhaustion in medical personnel due to the high number of cases, leading to more false positive predictions. Robust deep CNN models are capable of comprehending these complex images. The proposed work presents a novel hybrid ERCNet model that was designed by integrating the finest aspects of EfficientNetv2Small, ResNet50, and Capsule Network. The EfficientNetv2Small and ResNet50 models are concatenated to extract features from the multimodality images, while the capsule network is used to record the spatial hierarchies between features. The model is being trained and tested on multiple imaging modalities, including histopathology, mammography and ultrasound images. The model has performed well, attaining an average accuracy of 99.6% and an average loss of 0.012. Individual outcomes were also good, with histology accuracy of 100%, mammography accuracy of 99.7%, and ultrasound imaging accuracy of 99.1%, all with good precision, recall, and F1-score. All these results suggest that the proposed model ERCNet can not only be used in breast cancer diagnosis but can be applied in any medical imaging task in future.