EUVIMEDOpdag EUVIMED
Open access · CC BY 4.0

Nasal-HNet: A multimodal hierarchical network for joint anatomical and functional assessment of nasal obstruction from CBCT and clinical data

Qiang Wang, Xinyu Sui, Shihao Li, Shulin Cui, Xinran Zhang, Weibo Song

AIP Advances · 2026-11-01 · 16 (11)

DOI: 10.1063/5.0342200 ↗

At a glance

Journal article in AIP Advances, published online on 2026-11-17. Written by Qiang Wang and 5 co-authors. 32 references. The full text is freely available (open access, CC BY 4.0). Funded by: Provincial Key Specialty Research Project Fund, Dalian Municipal Guidance Program for Life and Health Sector.

AbstractBibliografiske oplysningerCiteRelated studies

Abstract (original language)

To address the longstanding discrepancy between patients’ subjective perception of nasal obstruction and objective clinical assessments, which makes it difficult to achieve accurate and quantitative evaluation of nasal airflow function, this study proposes an end-to-end multimodal deep learning framework for the precise prediction of subjective nasal obstruction scores (NOSE scores). The proposed model integrates heterogeneous medical data, including three-dimensional cone-beam computed tomography images, two-dimensional computational fluid dynamics airflow maps, and one-dimensional structured clinical variables. For feature extraction, a lightweight 3D convolutional neural network (3D-CNN) is designed to capture anatomical characteristics, while a dual-stream CNN-LSTM architecture is employed to model the spatiotemporal dynamics of airflow patterns. In addition, a novel LightGBM-based leaf embedding strategy is introduced to effectively encode discrete clinical features. Multimodal representations are then deeply fused through a Transformer-based cross-attention mechanism and a bidirectional gated recurrent unit. Finally, an improved sigmoid-based regression head constrains the output to a continuous range of 0–20, consistent with the NOSE scoring system. Experimental results demonstrate that the proposed model achieves a mean absolute error of 3.4668 on an independent test set and significantly outperforms single-modality baselines. These findings suggest that the model provides a feasible and clinically meaningful approach for the objective quantification of subjective nasal obstruction perception.

Funding

  • Provincial Key Specialty Research Project Fund
  • Dalian Municipal Guidance Program for Life and Health Sector

Cite

APA

Wang, Q., Sui, X., Li, S., Cui, S., Zhang, X., & Song, W. (2026). Nasal-HNet: A multimodal hierarchical network for joint anatomical and functional assessment of nasal obstruction from CBCT and clinical data. AIP Advances, 16(11). https://doi.org/10.1063/5.0342200

Vancouver

Wang Q, Sui X, Li S, Cui S, Zhang X, Song W. Nasal-HNet: A multimodal hierarchical network for joint anatomical and functional assessment of nasal obstruction from CBCT and clinical data. AIP Advances. 2026;16(11). doi:10.1063/5.0342200

BibTeX
@article{Wang2026,
  author = {Wang, Qiang and Sui, Xinyu and Li, Shihao and Cui, Shulin and Zhang, Xinran and Song, Weibo},
  title = {Nasal-HNet: A multimodal hierarchical network for joint anatomical and functional assessment of nasal obstruction from CBCT and clinical data},
  journal = {AIP Advances},
  year = {2026},
  volume = {16},
  number = {11},
  doi = {10.1063/5.0342200}
}
Start for free →