Scholarlink

<p>Pulmonary nodule detection plays a crucial role in computer-aided diagnosis systems. Thoracic computed tomography (CT) enables the diagnosis of lung cancer in a timely manner through the detection of pulmonary nodules. Convolutional neural networks (CNNs) have proven to be highly effective for this task. In this study, we propose TSC-Nodule, a novel two-stage pulmonary nodule detection framework designed to overcome key limitations of existing methods. Previous models often struggle to capture both global context and fine-grained local details simultaneously. To address this, we introduce the Tied Non-Local and Spatial-Channel Reconstructed Residual (TSCNR) module. This module integrates non-local operations, tied block convolutions, residual learning, and spatial-channel reconstruction, enhancing feature encoding by capturing long-range dependencies while preserving local structural cues. The TSCNR module is versatile and can be easily integrated into existing pulmonary nodule detection networks. TSC-Nodule employs a two-stage detection strategy incorporating false positive reduction and result fusion, thereby improving both sensitivity and specificity. Experiments on the publicly available LUNA16 dataset demonstrate that TSC-Nodule achieves superior accuracy, particularly for detecting small or ambiguous nodules, while significantly reducing false positives. The TSC-Nodule exceeds other methods in terms of the Competition Performance Metric (CPM), achieving a 0.82% improvement over DFNodule. The framework demonstrates significantly enhanced recognition performance at lower false positive rates and achieves the best performance at higher false positive rates. Furthermore, TSC-Nodule achieves performance improvements of 0.13% in AUC and 0.24% in AP, which were found to be statistically significant. The proposed TSC-Nodule accurately detects pulmonary nodules exhibiting various morphological variations on CT images. The source code is available at: <a href="https://github.com/bonaldo112/TSC-Nodule/TSC-Nodule/">https://github.com/bonaldo112/TSC-Nodule/TSC-Nodule/</a>.</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

TSC-Nodule: a tied non-local and spatial-channel reconstructed faster R-CNN for pulmonary nodule detection

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Ronald Bbosa,
  • Feng Liu,
  • Kafui Efio-Akolly,
  • Omar A. M. Salem,
  • Yi-Ping Phoebe Chen

摘要

Pulmonary nodule detection plays a crucial role in computer-aided diagnosis systems. Thoracic computed tomography (CT) enables the diagnosis of lung cancer in a timely manner through the detection of pulmonary nodules. Convolutional neural networks (CNNs) have proven to be highly effective for this task. In this study, we propose TSC-Nodule, a novel two-stage pulmonary nodule detection framework designed to overcome key limitations of existing methods. Previous models often struggle to capture both global context and fine-grained local details simultaneously. To address this, we introduce the Tied Non-Local and Spatial-Channel Reconstructed Residual (TSCNR) module. This module integrates non-local operations, tied block convolutions, residual learning, and spatial-channel reconstruction, enhancing feature encoding by capturing long-range dependencies while preserving local structural cues. The TSCNR module is versatile and can be easily integrated into existing pulmonary nodule detection networks. TSC-Nodule employs a two-stage detection strategy incorporating false positive reduction and result fusion, thereby improving both sensitivity and specificity. Experiments on the publicly available LUNA16 dataset demonstrate that TSC-Nodule achieves superior accuracy, particularly for detecting small or ambiguous nodules, while significantly reducing false positives. The TSC-Nodule exceeds other methods in terms of the Competition Performance Metric (CPM), achieving a 0.82% improvement over DFNodule. The framework demonstrates significantly enhanced recognition performance at lower false positive rates and achieves the best performance at higher false positive rates. Furthermore, TSC-Nodule achieves performance improvements of 0.13% in AUC and 0.24% in AP, which were found to be statistically significant. The proposed TSC-Nodule accurately detects pulmonary nodules exhibiting various morphological variations on CT images. The source code is available at: https://github.com/bonaldo112/TSC-Nodule/TSC-Nodule/.