CPM-XNet: Annotation-Efficient Deep-Learning Framework for Detecting Tuberculosis in Chest X-Ray Images.
Authors
Affiliations (10)
Affiliations (10)
- Department of Medical Imaging, Tungs' Taichung MetroHarbor Hospital, Taichung 435403, Taiwan.
- Doctoral Program in Medical Biotechnology, National Chung Hsing University, Taichung City 40227, Taiwan.
- Division of Thoracic Surgery, Department of Surgery, Changhua Christian Hospital, No. 135, Nanhsiao Street, Changhua County 500209, Taiwan.
- Department of Post-Baccalaureate Medicine, College of Medicine, National Chung Hsing University, No. 145, Xingda Road, South Dist., Taichung City 402202, Taiwan.
- Executive Master Program in Life Sciences, National Chung Hsing University, No. 145, Xingda Road, South Dist., Taichung City 402202, Taiwan.
- Department of Medical Research, Tungs' Taichung MetroHarbor Hospital, Taichung 435403, Taiwan.
- School of Medical Informatics, Chung Shan Medical University and IT Office, Chung Shan Medical University Hospital, Taichung 40201, Taiwan.
- Department of Information Management, Ming Chuan University, Taoyuan 33348, Taiwan.
- Graduate Institute of Genomics and Bioinformatics, National Chung Hsing University, Taichung City 40227, Taiwan.
- Institute of Molecular Biology, National Chung Hsing University, Taichung City 40227, Taiwan.
Abstract
<b>Background/Objectives</b>: Chest X-ray (CXR) images are a widely used first-line screening tool for pulmonary tuberculosis (TB) detection but are difficult to interpret, which has increased demand for an automated screening tool. Deep-learning-based computer-aided diagnosis systems have demonstrated a classification performance comparable to that of trained radiologists, but they rely on dense annotations such as lesion-level or pixel-level labels, which are costly and difficult to obtain in routine clinical workflows. We developed CPM-XNet, an annotation-efficient framework for lesion-annotation-free downstream TB classification in CXR images. <b>Methods</b>: CPM-XNet incorporates a compressing-projecting mask (CPM) to provide soft lung-aware modulation while preserving global contextual information. The CPM-modulated images are then used for downstream classification with multiple convolutional neural network backbones and a vision transformer baseline. <b>Results</b>: Experiments were conducted using an internal hospital dataset and public TB datasets, and CPM-XNet showed improved performance compared with baseline models trained on unmodulated images. In a repeated-seed evaluation of the main ResNet-101 configuration on the Tung cohort, CPM-ResNet101 showed higher and more stable performance than the non-CPM counterpart and demonstrated significant paired improvement using McNemar's exact test. An ablation analysis indicated that CPM modulation was the main contributor to performance improvement while data augmentation and the classifier architecture further influenced the overall robustness. <b>Conclusions</b>: CPM-XNet provides an annotation-efficient strategy for lesion-annotation-free downstream TB classification in CXR images. The findings support preliminary technical feasibility, although larger, naturally imbalanced, cross-institutional validation is required before clinical deployment can be inferred.