CDO-VIB DenseNet: An Improved DenseNet-121 Network for Multi-Label Chest X-ray Disease Classification
Health Medicine and Therapeutics · 2026-07-29 · 2 (1) · 65
DOI: 10.63313/hmt.9031 ↗At a glance
Journal article in Health Medicine and Therapeutics, published online on 2026-07-29. Written by Chenxin Pan. 10 references. The full text is freely available (open access, CC BY 4.0).
Abstract (original language)
Multi-label chest X-ray classification is challenging because multiple diseases may appear simultaneously, abnormal regions are often subtle, and class distributions are strongly imbalanced. This paper proposes CDO-VIB DenseNet, an improved DenseNet-121 network for ChestX-ray14 classification. The model introduces Dense Squeeze-and-Excitation channel recalibration, Asymmetric Convolution Blocks, and a Class-Decoupled Orthogonal Variational Information Bottleneck head with evidential prediction. Based on available ROC screenshots, the macro-average AUC improves from 0.79 for the DenseNet-121 baseline to 0.81 for the proposed model. The results indicate that disease-specific attention, variational representation compression, and orthogonal global-local fusion are effective for multi-label radiographic classification.
Cite
APA
Pan, C. (2026). CDO-VIB DenseNet: An Improved DenseNet-121 Network for Multi-Label Chest X-ray Disease Classification. Health Medicine and Therapeutics, 2(1), 65. https://doi.org/10.63313/hmt.9031
Vancouver
Pan C. CDO-VIB DenseNet: An Improved DenseNet-121 Network for Multi-Label Chest X-ray Disease Classification. Health Medicine and Therapeutics. 2026;2(1):65. doi:10.63313/hmt.9031
BibTeX
@article{Pan2026,
author = {Pan, Chenxin},
title = {CDO-VIB DenseNet: An Improved DenseNet-121 Network for Multi-Label Chest X-ray Disease Classification},
journal = {Health Medicine and Therapeutics},
year = {2026},
volume = {2},
number = {1},
pages = {65},
doi = {10.63313/hmt.9031}
}
EUVIMED