Preliminary Evaluation of a Machine Learning Model for Detecting Mediastinal Abnormalities on Frontal Chest Radiographs
Main Article Content
Abstract
SUMMARY
Objective: To preliminarily evaluate the feasibility and performance of a machine learning model in detecting mediastinal abnormalities on frontal chest radiographs.
Methods: A retrospective cross-sectional study was conducted at the University Medical Center Ho Chi Minh City from September 2018 to June 2025. The study dataset comprised 735 frontal chest radiographs, including 441 cases with mediastinal abnormalities (such as 260 cases of mediastinal tumors and 181 cases of thoracic aortic aneurysms), and 294 normal cases. All cases were correlated with chest computed tomography or magnetic resonance imaging to determine the presence or absence of mediastinal abnormalities. The chest radiographs were preprocessed to standardize image quality before being used to train a ResNet34 model using a supervised learning approach. After training, the model received frontal chest radiographs as input and classified them into two groups: normal and abnormal mediastinum. The diagnostic performance of the model was evaluated using five-fold cross-validation.
Results: In the mediastinal lesion group, males predominated at 59.2%, and the mean age was significantly higher than that of the non-lesion group (58.4 ± 19.1 versus 43.4 ± 14.2 years; p<0.05). The difference in gender distribution between the two groups was statistically significant (p<0.05). The two most common radiographic signs in the lesion group were abnormal mediastinal contour (66.7%) and mediastinal widening (30.4%). Regarding the diagnostic performance, the ResNet34 model achieved an area under the curve (AUC-ROC) of 0.95 (95% CI: 0.93-0.98) and an accuracy of 0.87 (95% CI: 0.82-0.92). Other performance metrics included an F1-score of 0.88 (95% CI: 0.82-0.94), sensitivity of 0.83 (95% CI: 0.72-0.94), specificity of 0.93 (95% CI: 0.88-0.98), positive predictive value of 0.95 (95% CI: 0.92-0.98), and negative predictive value of 0.79 (95% CI: 0.69-0.88).
Conclusion: Preliminary results demonstrate that the machine learning model has high diagnostic performance in detecting mediastinal abnormalities on frontal chest radiographs.
Keywords: machine learning model, chest radiograph, mediastinal lesion.
Keywords
machine learning model, chest radiograph, mediastinal lesion.
Article Details
References
Am J Med. 2004;116(2):73-7. doi:10.1016/j.amjmed.2003.08.030
2. Wernecke K, Vassallo P, Pötter R, Lückener HG, Peters PE. Mediastinal tumors: sensitivity of detection with sonography compared with CT and radiography. Radiology. 1990;175(1):137-43. doi:10.1148/radiology.175.1.2179987
3. Rehman A, Khan A, Fatima G, Naz S, Razzak I. Review on chest pathologies detection systems using deep
learning techniques. Artif Intell Rev. 2023:1-47. doi:10.1007/s10462-023-10457-9
4. He K, Zhang X, Ren S, Sun J. Deep Residual Learning for Image Recognition. 2016:770-778.
5. Pydicom: An open-source DICOM library for Python. Zenodo; 2024. https://doi.org/10.5281/zenodo.13824606
6. Harris CR, Millman KJ, van der Walt SJ, Gommers R, Virtanen P. Array programming with NumPy. Nature.
2020;585(7825):357-362. doi:10.1038/s41586-020-2649-2
7. Bradski G. The OpenCV Library. Dr Dobb’s Journal of Software Tools. 2000;25(11):120-125.
8. Van der Walt S, Schönberger JL, Nunez-Iglesias J, Boulogne F, Warner JD. scikit-image: image processing in
Python. PeerJ. 2014;2:e453. doi:10.7717/peerj.453
9. Paszke A, Gross S, Massa F, Lerer A, Bradbury J. PyTorch: An Imperative Style, High-Performance Deep Learning Library.
presented at: Advances in Neural Information Processing Systems 32; 2019; https://doi.org/10.48550/arXiv.1912.01703
10. Marcel S, Rodriguez Y. Torchvision the machine-vision package of torch. presented at: Proceedings of the 18th
ACM International Conference on Multimedia; 2010; https://doi.org/10.1145/1873951.1874254
11. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B. Scikit-learn: Machine Learning in Python. Journal of
Machine Learning Research. 2011;12:2825-2830.
12. Zuiderveld K. Contrast Limited Adaptive Histogram Equalization. In: Heckbert PS, ed. Graphics Gems IV. Academic
Press; 1994:474-485.
13. Faiza Z, Sharman T. Thoracic Aorta Aneurysm. StatPearls [Internet]. StatPearls Publishing; 2026.
14. Mehrabi Nasab E, Athari SS. The prevalence of thoracic aorta aneurysm as an important cardiovascular disease
in the general population. J Cardiothorac Surg. 2022;17(1):51. doi:10.1186/s13019-022-01767-0
15. Kai C, Kasai S, Teramoto R, et al. Classifying abnormalities in chest radiographs from Vietnam using deep learning
for early detection of cardiopulmonary diseases. Frontiers in Radiology. 2025;doi:10.3389/fradi.2025.1703927
16. Nguyen NH, Nguyen HQ, Nguyen NT, et al. Deployment and validation of an AI system for detecting abnormal
chest radiographs in clinical settings. Front Digit Health. 2022;4:890759. doi:10.3389/fdgth.2022.890759
17. Ribeiro E, Cardenas DAC, Dias FM, Krieger JE, Gutierrez MA. Explainable artificial intelligence in deep learningbased
detection of aortic elongation on chest X-ray images. Eur Heart J Digit Health. 2024;5(5):524-534.
doi:10.1093/ehjdh/ztae045
18. Van Beek EJR, Ahn JS, Kim MJ, Murchison JT. Validation study of machine-learning chest radiograph software in
primary and emergency medicine. Clin Radiol. 2023;78(1):1-7. doi:10.1016/j.crad.2022.08.129