Journal article
Classifying chronic obstructive pulmonary disease status using computed tomography imaging and convolutional neural networks: comparison of model input image types and training data severity
Journal of medical imaging (Bellingham, Wash.), Vol.12(3), 034502
05/2025
DOI: 10.1117/1.JMI.12.3.034502
PMCID: PMC12097752
PMID: 40415865
Abstract
Convolutional neural network (CNN)-based models using computed tomography images can classify chronic obstructive pulmonary disease (COPD) with high performance, but various input image types have been investigated, and it is unclear what image types are optimal. We propose a 2D airway-optimized topological multiplanar reformat (tMPR) input image and compare its performance with established 2D/3D input image types for COPD classification. As a secondary aim, we examined the impact of training on a dataset with predominantly mild COPD cases and testing on a more severe dataset to assess whether it improves generalizability.
CanCOLD study participants were used for training/internal testing; SPIROMICS participants were used for external testing. Several 2D/3D input image types were adapted from the literature. In the proposed models, 2D airway-optimized tMPR images (to convey shape and interior/contextual information) and 3D output fusion of axial/sagittal/coronal images were investigated. The area-under-the-receiver-operator-curve (AUC) was used to evaluate model performance and Brier scores were used to evaluate model calibration. To further examine how training dataset severity impacts generalization, we compared model performance when trained on the milder CanCOLD dataset versus the more severe SPIROMICS dataset, and vice versa.
A total of
CanCOLD participants were used for training/validation and
for testing;
SPIROMICS participants were used for external testing. For the CanCOLD and SPIROMICS test set, the proposed 2D tMPR on its own (CanCOLD:
; SPIROMICS:
) and combined with the 3D axial/coronal/sagittal lung view (CanCOLD:
; SPIROMICS:
) had the highest performance. The combined 2D tMPR and 3D axial/coronal/sagittal lung view had the lowest Brier score (CanCOLD: score = 0.16; SPIROMICS: score = 0.24). Conversely, using SPIROMICS for training/testing and CanCOLD for external testing resulted in lower performance when tested on CanCOLD for 2D tMPR on its own (SPIROMICS: AUC = 0.92; CanCOLD: AUC = 0.74) and when combined with the 3D axial/coronal/sagittal lung view (SPIROMICS:
; CanCOLD:
).
The CNN-based model with the combined 2D tMPR images and 3D lung view as input image types had the highest performance for COPD classification, highlighting the importance of airway information and that the fusion of different types of information as input image can improve CNN-based model performance. In addition, models trained on CanCOLD demonstrated strong generalization to the more severe SPIROMICS cohort, whereas training on SPIROMICS resulted in lower performance when tested on CanCOLD. These findings suggest that training on milder COPD cases may improve classification performance across the disease spectrum.
Details
- Title: Subtitle
- Classifying chronic obstructive pulmonary disease status using computed tomography imaging and convolutional neural networks: comparison of model input image types and training data severity
- Creators
- Sara Rezvanjou - Toronto Metropolitan UniversityAmir Moslemi - Sunnybrook HospitalSamuel Peterson - Vida DiagnosticsWan-Cheng Tan - University of British ColumbiaJames C Hogg - University of British ColumbiaJean Bourbeau - Royal Victoria Regional Health CentreJoseph M Reinhardt - University of IowaMiranda Kirby - Toronto Metropolitan University
- Resource Type
- Journal article
- Publication Details
- Journal of medical imaging (Bellingham, Wash.), Vol.12(3), 034502
- DOI
- 10.1117/1.JMI.12.3.034502
- PMID
- 40415865
- PMCID
- PMC12097752
- NLM abbreviation
- J Med Imaging (Bellingham)
- ISSN
- 2329-4302
- eISSN
- 2329-4310
- Publisher
- SPIE-SOC PHOTO-OPTICAL INSTRUMENTATION ENGINEERS
- Grant note
- National Heart, Lung, and Blood Institute (NHLBI)Keck School of Medicine of USC
The authors would also like to thank the men and women who participated in the study and individuals in the *CanCOLD Collaborative research Group: Jonathon Samet (the Keck School of Medicine of USC, California, United States); Milo Puhan (John Hopkins School of Public Health, Baltimore, United States); Qutayba Hamid, Carolyn Baglole, Palmina Mancino, Pei-Zhi Li, Zhi Song, Dennis Jensen, Benjamin Mcdonald Smith (McGill University, Montreal, Quebec, Canada); Yvan Fortier and Mina Dligui (Sherbrooke University, Sherbrooke, Quebec, Canada); Kenneth Chapman, Jane Duke, Andrea S Gershon, Teresa To, (University of Toronto, Toronto, Ontario, Canada); J Mark Fitzgerald, Mohsen Sadatsafavi (University of British Columbia, Vancouver, Canada); Christine Lo, Sarah Cheng, Elena Un, Michael Cheng, Cynthia Fung, Nancy Haynes, Liyun Zheng, LingXiang Zou, Joe Comeau, Jonathon Leipsic, Cameron Hague (UBC James Hogg Research Center, Vancouver, Canada); Brandie L Walker, Curtis Dumonceaux, (University of Calgary, Calgary, Alberta, Canada); Paul Hernandez, Scott Fulton, (University of Dalhousie, Halifax, Nova Scotia, Canada); Shawn Aaron, Kathy Vandemheen, (University of Ottawa, Ottawa, Ontario, Canada); Denis O'Donnell, Matthew McNeil, Kate Whelan (Queen's University, Kingston, Ontario, Canada); Francois Maltais, Cynthia Brouillard (University of Laval, Quebec City, Quebec, Canada); Darcy Marciniuk, Ron Clemens, Janet Baran (University of Saskatchewan, Saskatoon, Saskatchewan, Canada).
- Language
- English
- Date published
- 05/2025
- Academic Unit
- Roy J. Carver Department of Biomedical Engineering; Radiology
- Record Identifier
- 9984824228502771
Metrics
34 Record Views