Göreve özgü optimize edilmiş bir CNN kullanılarak göğüs radyografilerinden etkin ve açıklanabilir pnömokonyoz taraması

Öz

Arka Plan—Pnömokonyoz, göğüs radyografilerinde ince, düşük kontrastlı ve yaygın olarak dağılmış parankimal anormalliklerle karakterize kronik bir mesleki akciğer hastalığıdır. Bu görüntüleme özellikleri, sınırlı veri seti mevcudiyeti ve ciddi sınıf dengesizliği ile birlikte güvenilir otomatik tespiti güçleştirmektedir. Son yıllarda derin öğrenme yaklaşımları umut verici sonuçlar göstermiş olmasına rağmen bu yöntemler çoğunlukla giderek daha karmaşık ve hesaplama açısından maliyetli mimarilere dayanmakta; ağ tasarımı ile hastalığa özgü radyografik örüntüler arasındaki uyum ise yeterince incelenmemektedir.

Amaç—Bu çalışmanın amacı, posteroanterior göğüs radyografilerinden pnömokonyozun otomatik tespiti için etkin ve göreve özgü bir derin öğrenme çerçevesi geliştirmektir. Temel amaç, kompakt ve optimize edilmiş bir evrişimsel mimarinin, aşırı model karmaşıklığına ihtiyaç duymadan rekabetçi performans sağlayıp sağlayamayacağını ve tarama uygulamaları için gerekli duyarlılığı koruyup koruyamayacağını araştırmaktır.

Yöntem—ResPneumoNet olarak adlandırılan hafif bir evrişimsel sinir ağı önerilmiştir. Bu yapıda ağ derinliği ve evrişimsel çekirdek boyutu, göreve bağlı mimari hiperparametreler olarak ele alınmış ve pnömokonyoza özgü radyografik örüntülere uygun alıcı alan ve özellik çıkarım kapasitesini belirlemek amacıyla Bayesçi hiperparametre optimizasyonu ile birlikte optimize edilmiştir. Sınıf dengesizliğini gidermek amacıyla, anatomik ve dokusal tutarlılığı koruyan dengeli bir veri seti oluşturmak için difüzyon tabanlı sentetik görüntü üretimi (MedFusion) kullanılmıştır. Model, 224×224 piksele yeniden boyutlandırılan göğüs radyografileri üzerinde eğitilmiş ve standart sınıflandırma ölçütleri ile ROC analizi kullanılarak değerlendirilmiştir. Ayrıca modelin açıklanabilirliğini değerlendirmek amacıyla akciğer bölgeleriyle sınırlandırılmış Grad-CAM analizi gerçekleştirilmiştir.

Bulgular—112 normal ve 28 pnömokonyoz-pozitif radyografiden oluşan bağımsız gerçek test setinde ResPneumoNet %76.43 doğruluk, %64.29 duyarlılık, %79.46 özgüllük ve 0.752 AUC değerine ulaşmış ve yalnızca 107.1k eğitilebilir parametre içermiştir. Daha derin referans modeller daha yüksek genel ayırt etme performansı göstermiştir; bu durum ResPneumoNet’in maksimum sınıflandırma başarısından ziyade mimari kompaktlığı önceliklendirdiğini göstermektedir. Akciğer bölgeleriyle sınırlandırılmış Grad-CAM analizlerinde model dikkatinin genel olarak akciğer alanlarında yoğunlaştığı gözlenmiş ve anatomik açıdan anlamlı karar davranışına yönelik nitel destek elde edilmiştir.

Sonuç—Bulgular, göreve uyumlu mimari optimizasyonun pnömokonyoz tespiti için kompakt ve hesaplama açısından etkin bir alternatif sağlayabileceğini göstermektedir. Önerilen çerçeve genel ayırt etme performansı açısından daha derin referans modelleri aşmamış olmakla birlikte, düşük model karmaşıklığı ile tarama odaklı duyarlılık arasında bir denge sunmaktadır. Gelecek çalışmalarda farklı veri setlerinde dış doğrulama yapılması ve yöntemin hastalık şiddetini dikkate alan analizlere genişletilmesi hedeflenmektedir.

Anahtar Kelimeler

• Bayesçi optimizasyon
• Göğüs radyografisi
• Difüzyon tabanlı veri artırma
• Hafif CNN
• Mesleki sağlık
• Pnömokonyoz

Destekleyen Kurum

Türkiye Bilim ve Teknolojik Araştırma Kurumu (TÜBİTAK) 2209-A Araştırma Programı

Proje Numarası

1919B012466614

Etik Beyan

Bu makale için etik kurul onayı gerekmemektedir.

Teşekkür

Yazarlar, Türkiye Bilimsel ve Teknolojik Araştırma Kurumu (TÜBİTAK) tarafından 2209-A Araştırma Programı (Proje No: 1919B012466614) kapsamında sağlanan destek için teşekkür eder.

Efficient and explainable pneumoconiosis screening from chest radiographs using a task-optimized CNN

Öz

Context—Pneumoconiosis is a chronic occupational lung disease characterized by subtle, low-contrast, and diffusely distributed parenchymal abnormalities on chest radiographs. These imaging characteristics, combined with limited dataset availability and severe class imbalance, make reliable automated detection challenging. Although recent deep learning approaches have shown promising results, they often rely on increasingly complex and computationally demanding architectures, while the alignment between network design and disease-specific radiographic patterns remains insufficiently explored.

Objective—This study aims to develop an efficient and task-specific deep learning framework for the automatic detection of pneumoconiosis from posteroanterior chest radiographs. The primary objective is to investigate whether a compact and optimized convolutional architecture can achieve competitive performance without relying on excessive model complexity, while maintaining high sensitivity required for screening applications.

Method—A lightweight convolutional neural network, termed ResPneumoNet, is proposed, where network depth and convolutional kernel size are treated as task-dependent architectural hyperparameters. These parameters are jointly optimized using Bayesian hyperparameter optimization to align the receptive field and feature extraction capacity with pneumoconiosis-specific radiographic patterns. To address class imbalance, diffusion-based synthetic image generation (MedFusion) is employed to construct a balanced dataset while preserving anatomical and textural consistency. The model is trained and evaluated on chest radiographs resized to 224×224 pixels, using standard classification metrics and ROC analysis. Additionally, lung-region–constrained Grad-CAM is utilized to assess model interpretability.

Results—On the independent real test set consisting of 112 normal and 28 pneumoconiosis-positive radiographs, ResPneumoNet achieved an accuracy of 76.43%, sensitivity of 64.29%, specificity of 79.46%, and an AUC of 0.752, with 107.1k trainable parameters. Deeper reference models achieved higher overall discriminative performance, indicating that ResPneumoNet prioritizes architectural compactness over maximum classification accuracy. Lung-region–constrained Grad-CAM analyses showed that model attention was generally concentrated within lung-field regions, providing qualitative support for anatomically relevant decision behavior.

Conclusion—The findings indicate that task-aligned architectural optimization can provide a compact and computationally efficient alternative for pneumoconiosis detection. Although the proposed framework did not outperform deeper reference models in overall discrimination, it offered reduced model complexity with screening-oriented sensitivity. Future work will focus on external validation across diverse datasets and extending the approach toward severity-aware analysis.

Anahtar Kelimeler

• Bayesian optimization
• Chest radiography
• Diffusion-based augmentation
• Lightweight CNN
• Occupational health
• Pneumoconiosis

Destekleyen Kurum

The Scientific and Technological Research Council of Turkey (TÜBİTAK) 2209-A Research Program

Proje Numarası

1919B012466614

Etik Beyan

Ethics committee approval is not required for this article.

Teşekkür

The authors gratefully acknowledge the support provided by the Scientific and Technological Research Council of Turkey (TÜBİTAK) under the 2209-A Research Program (Project No. 1919B012466614).

Kaynakça

1. Y. Zhang, W. Xuan, S. Chen, M. Yang, H. Xing, “The Screening and Diagnosis Technologies Towards Pneumoconiosis: From Imaging Analysis to E-Noses”, Chemosensors, 13(3), 102, 2025. https://doi.org/10.3390/chemosensors13030102.
2. C. C. Leung, I. T. S. Yu, W. Chen, “Silicosis”, Lancet, 379(9830), 2008–2018, 2012. https://doi.org/10.1016/S0140-6736(12)60235-9.
3. S. Chong, K. S. Lee, M. J. Chung, J. Han, O. J. Kwon, T. S. Kim, “Pneumoconiosis: comparison of imaging and pathologic findings”, Radiographics, 26(1), 59–77, 2006. https://doi.org/10.1148/rg.261055070.
4. P. Cullinan, et al., “Occupational lung diseases: from old and novel exposures to effective preventive strategies”, The Lancet Respiratory Medicine, 5(3), 445–455, 2017. https://doi.org/10.1016/S2213-2600(16)30424-6.
5. M. Akgün, B. Ergan, “Silicosis in Turkey: Is it an Endless Nightmare or is There Still Hope?”, Turkish Thoracic Journal, 19(2), 89–93, 2018. https://doi.org/10.5152/TurkThoracJ.2018.040189.
6. Y. Chen, D. Liu, H. Ji, W. Li, Y. Tang, “Global and regional burden of pneumoconiosis, 1990–2021: an analysis of data from the global burden of disease study 2021”, Frontiers in Medicine, 12, 1559540, 2025. https://doi.org/10.3389/fmed.2025.1559540.
7. A. W. Matyga, L. Chelala, J. H. Chung, “Occupational Lung Diseases: Spectrum of Common Imaging Manifestations”, Korean Journal of Radiology, 24(8), 795–806, 2023. https://doi.org/10.3348/kjr.2023.0274.
8. C. W. Cox, C. S. Rose, D. A. Lynch, “State of the Art: Imaging of Occupational Lung Disease”, Radiology, 270(3), 681–696, 2014. https://doi.org/10.1148/radiol.13121415.

9. J. S. Jun, J. I. Jung, H. R. Kim, M. I. Ahn, D. H. Han, J. M. Ko, S. H. Park, H. G. Lee, H. Arakawa, J. W. Koo, “Complications of pneumoconiosis: Radiologic overview”, European Journal of Radiology, 82(10), 1819–1830, 2013. https://doi.org/10.1016/j.ejrad.2013.05.026.
10. Guidelines for the Use of the ILO International Classification of Radiographs of Pneumoconioses, Revised edition 2022, International Labour Organization, Geneva, Switzerland, 2022.
11. C. N. Halldin, D. J. Blackley, E. L. Petsonk, A. S. Laney, “Pneumoconioses Radiographs in a Large Population of U.S. Coal Workers: Variability in A Reader and B Reader Classifications by Using the International Labour Office Classification”, Radiology, 284(3), 870–876, 2017. https://doi.org/10.1148/radiol.2017162437.
12. W. Sun, D. Wu, Y. Luo, L. Liu, H. Zhang, S. Wu, Y. Zhang, C. Wang, H. Zheng, J. Shen, C. Luo, “ExpertNet: Defeat noisy labels by deep expert consultation paradigm for pneumoconiosis staging on chest radiographs”, Expert Systems with Applications, 232, 120710, 2023. https://doi.org/10.1016/j.eswa.2023.120710.
13. W. Xue, Y. Gao, X. Ren, J. Cui, J. Zhao, Y. Qiang, Y. Wei, Y. Qiao, “WIDINet: A diagnostic model for staging pneumoconiosis based on data expansion and KL entropy judgement”, Biomedical Signal Processing and Control, 90, 105741, 2024. https://doi.org/10.1016/j.bspc.2023.105741.
14. Q. Wei, X. Li, D. Liu, Z. Liao, “MFTAnet: Two-step Aggregation Net of Multiscale Features for Pneumoconiosis Screening”, Proceedings of the 6th ACM International Conference on Multimedia in Asia (MMASIA 24), Auckland, New Zealand, 03–06 December 2024. https://doi.org/10.1145/3696409.3700274.
15. M. S. Alam, D. Wang, A. Sowmya, “DLA-Net: dual lesion attention network for classification of pneumoconiosis using chest X-ray images”, Scientific Reports, 14(1), 11616, 2024. https://doi.org/10.1038/s41598-024-61024-3.
16. X. Ren, S. Chu, G. Ji, Z. Zhao, J. Zhao, Y. Qiang, Y. Wei, Y. Wang, “OMSF2: optimizing multi-scale feature fusion learning for pneumoconiosis staging diagnosis through data specificity augmentation”, Complex and Intelligent Systems, 11(1), 109, 2025. https://doi.org/10.1007/s40747-024-01729-0.
17. M. S. Alam, D. Wang, A. Sowmya, “AMFP-net: Adaptive multi-scale feature pyramid network for diagnosis of pneumoconiosis from chest X-ray images”, Artificial Intelligence in Medicine, 154, 102917, 2024. https://doi.org/10.1016/j.artmed.2024.102917.
18. Q. Wei, W. Song, L. Fu, Y. Lei, Q. Wang, “Digital Radiography-Based Pneumoconiosis Diagnosis via Vision Transformer Networks”, Journal on Artificial Intelligence, 7, 39–53, 2025. https://doi.org/10.32604/jai.2025.063188.
19. Y. Huang, Y. Si, B. Hu, Y. Zhang, S. Wu, D. Wu, Q. Wang, “Transformer-based factorized encoder for classification of pneumoconiosis on 3D CT images”, Computers in Biology and Medicine, 150, 106137, 2022. https://doi.org/10.1016/j.compbiomed.2022.106137.
20. Y. Liu, J. Wu, J. Zhou, J. Guo, C. Liang, Y. Xing, Z. Wang, L. Chen, Y. Ding, D. Ren, Y. Bai, D. Hu, “Identification of high-risk population of pneumoconiosis using deep learning segmentation of lung 3D images and radiomics texture analysis”, Computer Methods and Programs in Biomedicine, 244, 108006, 2024. https://doi.org/10.1016/j.cmpb.2024.108006.
21. Z. Zhou, X. Li, H. Ji, X. Xu, Z. Chang, K. Wu, Y. Song, M. Kao, H. Chen, D. Wu, T. Zhang, “Application of improved Unet network in the recognition and segmentation of lung CT images in patients with pneumoconiosis”, BMC Medical Imaging, 24(1), 220, 2024. https://doi.org/10.1186/s12880-024-01377-3.
22. P. Belim, N. Bhatt, “Predicting Ceramic Workers’ Lung Health via X-ray Using Machine Learning”, 2025 International Conference on Artificial Intelligence and Machine Vision (AIMV), Gandhinagar, India, 16–17 August 2025. https://doi.org/10.1109/AIMV66517.2025.11203746.
23. C. Hao, N. Jin, C. Qiu, K. Ba, X. Wang, H. Zhang, Q. Zhao, B. Huang, “Balanced Convolutional Neural Networks for Pneumoconiosis Detection”, International Journal of Environmental Research and Public Health, 18(17), 9091, 2021. https://doi.org/10.3390/ijerph18179091.
24. G. Müller-Franzes, J. M. Niehues, F. Khader, S. T. Arasteh, C. Haarburger, C. Kuhl, T. Wang, T. Han, T. Nolte, S. Nebelung, J. N. Kather, D. Truhn, “A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial networks for medical image synthesis”, Scientific Reports, 13(1), 12098, 2023. https://doi.org/10.1038/s41598-023-39278-0.
25. K. He, X. Zhang, S. Ren, J. Sun, “Deep Residual Learning for Image Recognition”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 27–30 June 2016, 770–778. https://doi.org/10.1109/CVPR.2016.90.
26. M. A. Gelbart, J. Snoek, R. P. Adams, “Bayesian Optimization with Unknown Constraints”, arXiv, https://doi.org/10.48550/arXiv.1403.5607.
27. J. Wu, X.-Y. Chen, H. Zhang, L.-D. Xiong, H. Lei, S.-H. Deng, “Hyperparameter Optimization for Machine Learning Models Based on Bayesian Optimization”, Journal of Electronic Science and Technology, 17(1), 26–40, 2019. https://doi.org/10.11989/JEST.1674-862X.80904120.
28. B. Shahriari, K. Swersky, Z. Wang, R. P. Adams, N. de Freitas, “Taking the Human Out of the Loop: A Review of Bayesian Optimization”, Proceedings of the IEEE, 104(1), 148–175, 2016. https://doi.org/10.1109/JPROC.2015.2494218.
29. A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Dollár, R. Girshick, “Segment Anything”, Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, 02–06 October 2023, 3992–4003. https://doi.org/10.1109/ICCV51070.2023.00371.
30. C. Barber, D. Fishwick, “Pneumoconiosis”, Medicine, 44(6), 355–358, 2016. https://doi.org/10.1016/j.mpmed.2016.03.001.
31. L. Fan, Z. Wang, J. Zhou, “LDADN: a local discriminant auxiliary disentangled network for key-region-guided chest X-ray image synthesis augmented in pneumoconiosis detection”, Biomedical Optics Express, 13(8), 4353–4369, 2022. https://doi.org/10.1364/BOE.461888.
32. Y. Zhang, B. Zheng, F. Zeng, X. Cheng, T. Wu, Y. Peng, Y. Zhang, Y. Xie, W. Yi, W. Chen, J. Wu, L. Li, “Potential of digital chest radiography-based deep learning in screening and diagnosing pneumoconiosis: An observational study”, Medicine, 103(25), e38478, 2024. https://doi.org/10.1097/MD.0000000000038478.
33. C. Zhang, J. He, L. Shang, “An X-ray image classification method with fine-grained features for explainable diagnosis of pneumoconiosis”, Personal and Ubiquitous Computing, 28(2), 403–415, 2024. https://doi.org/10.1007/s00779-023-01730-3.
34. S. Su, Y. Wang, Y. Zhu, Y. Dai, Z. Yu, Z.-R. Tang, B. Li, S. Wang, “MBSM-Net: A Multi-Branch Structure Model for Pneumoconiosis Screening and Grading of Chest X-Ray Images”, IET Image Processing, 19(1), e70128, 2025. https://doi.org/10.1049/ipr2.70128.
35. W. Li, T. Zhou, A. Dong, L. Xiong, Q. Luo, L. Mou, X. Liu, “Highly Accurate Occupational Pneumoconiosis Staging via Dark Channel Prior-Inspired Lesion Area Enhancement”, Journal of Imaging Informatics in Medicine, 38(6), 3886–3898, 2025. https://doi.org/10.1007/s10278-025-01472-z.
36. H. Dong, B. Zhu, X. Kong, X. Su, T. Liu, X. Zhang, “Deep learning-based algorithm for classifying high-resolution computed tomography features in coal workers’ pneumoconiosis”, Biomedical Engineering Online, 24(1), 7, 2025. https://doi.org/10.1186/s12938-025-01333-4.
37. Y. Ma, F. Cui, Y. Yao, F. Shen, H. Qin, B. Li, Y. Wang, “Machine learning models for the prediction of preclinical coal workers’ pneumoconiosis: integrating CT radiomics and occupational health surveillance records”, Journal of Translational Medicine, 23(1), 896, 2025. https://doi.org/10.1186/s12967-025-06907-3.
38. Q. Huang, H. Bai, H. Pan, W. Jiang, B. Liu, Y. Zhang, J. He, L. Li, “Early pneumoconiosis recognition from CT via progressive lesion awareness and multi-axis denoising attention mechanisms”, Pattern Recognition, 172, 112712, 2026. https://doi.org/10.1016/j.patcog.2025.112712.