Keywords
Pediatric Pneumonia
Medical Imaging
DA2C2f Module
DySample Module
Categories
How to Cite
Abstract
Early diagnosis of pediatric pneumonia is critical for reducing mortality rates, and automated detection based on chest X-rays is key to improving diagnostic efficiency. While deep learning-based object detection algorithms have shown promise in this task, existing methods often suffer from limited generalization and detection risks due to model complexity or loss of detail during feature extraction and upsampling processes. In this paper, we propose an enhanced YOLOv12n model specifically designed for pediatric pneumonia detection. We introduce the DA2C2f module, which leverages multi-level feature processing and deep convolution operations to improve the model's ability to capture complex features at different levels, effectively enhancing classification performance. Additionally, the DySample module utilizes dynamic upsampling to adaptively adjust the sampling points based on the local content of the input feature map, overcoming the limitations of traditional fixed upsampling methods and reducing image blurring and detail loss. Experimental results demonstrate that the proposed improved model outperforms competing models across key metrics, including recall rate, mean average precision (mAP), and model generalization. Specifically, our model achieves a 1.7% improvement in recall rate (85.4%), a 0.4% improvement in mAP@0.5 (84.4%), and a 0.4% improvement in mAP@0.5-0.95 (52.9%). Notably, in bacterial pneumonia detection, our model exhibits the highest average precision (AP) of 86.8%, a 2.2% improvement over YOLOv12n. Ablation studies further validate the critical role of the DA2C2f and DySample modules in enhancing model performance, particularly in improving recall rate while maintaining high detection accuracy. The improved YOLOv12n model holds significant potential for automated pediatric pneumonia diagnosis and provides valuable insights for future medical image analysis applications.
References
[1] Mahtab, S., Blau, D. M., Madewell, Z. J., Ogbuanu, I., Ojulong, J., Lako, S., ... & Swart, P. J. (2024). Post-mortem investiga-tion of deaths due to pneumonia in children aged 1–59 months in sub-Saharan Africa and South Asia from 2016 to 2022: an ob-servational study. The Lancet Child & Ado-lescent Health, 8(3), 201-213.
[2] Guo, K., Cheng, J., Li, K., Wang, L., Lv, Y., & Cao, D. (2023). Diagnosis and detec-tion of pneumonia using weak-label based on X-ray images: a multi-center study. BMC Medical Imaging, 23(1), 209.
[3] Mienye, I. D., Swart, T. G., Obaido, G., Jordan, M., & Ilono, P. (2025). Deep con-volutional neural networks in medical im-age analysis: A review. Information, 16(3), 195.
[4] Redmon J. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
[5] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 7263-7271.
[6] Farhadi A, Redmon J. Yolov3: An incre-mental improvement[C]//Computer vision and pattern recognition. Berlin/Heidelberg, Germany: Springer, 2018, 1804: 1-6.
[7] Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of ob-ject detection[J]. arxiv preprint arxiv:2004.10934, 2020.
[8] Zhang Y, Guo Z, Wu J, et al. Real-time ve-hicle detection based on improved yolo v5[J]. Sustainability, 2022, 14(19): 12274.
[9] Li C, Li L, Jiang H, et al. YOLOv6: A sin-gle-stage object detection framework for industrial applications[J]. arxiv preprint arxiv:2209.02976, 2022.
[10] Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object de-tectors[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023: 7464-7475.
[11] Wang G, Chen Y, An P, et al. UAV-YOLOv8: A small-object-detection model based on improved YOLOv8 for UAV aeri-al photography scenarios[J]. Sensors, 2023, 23(16): 7190.
[12] Wang C Y, Yeh I H, Mark Liao H Y. Yolov9: Learning what you want to learn using programmable gradient infor-mation[C]//European conference on com-puter vision. Cham: Springer Nature Swit-zerland, 2024: 1-21.
[13] Wang A, Chen H, Liu L, et al. Yolov10: Real-time end-to-end object detection[J]. arxiv preprint arxiv:2405.14458, 2024.
[14] Khanam R, Hussain M. Yolov11: An over-view of the key architectural enhance-ments[J]. arxiv preprint arxiv:2410.17725, 2024.
[15] Chen, Yiyun, et al. "Deep learning for clas-sification of pediatric chest radiographs by WHO’s standardized methodology." PLoS one 16.6 (2021): e0253239.
[16] Kundu, Rohit, et al. "Pneumonia detection in chest X-ray images using an ensemble of deep learning models." PloS one 16.9 (2021): e0256630.
[17] Radočaj, Petra, and Goran Martinović. "In-terpretable Deep Learning for Pediatric Pneumonia Diagnosis Through Multi-Phase Feature Learning and Activation Patterns." Electronics 14.9 (2025): 1899.
[18] Habib, Nahida, et al. "Ensemble of CheXNet and VGG-19 feature extractor with random forest classifier for pediatric pneumonia detection." SN Computer Sci-ence 1.6 (2020): 359.
[19] Yao, Dan, et al. "Accurate and intelligent diagnosis of pediatric pneumonia using X-ray images and blood testing data." Fron-tiers in bioengineering and biotechnology 11 (2023): 1058888.
[20] Kaya, Mahir. "Feature fusion-based ensem-ble CNN learning optimization for auto-mated detection of pediatric pneumonia." Biomedical Signal Processing and Control 87 (2024): 105472.
[21] Xin, Kevin Z., David Li, and Paul H. Yi. "Limited generalizability of deep learning algorithm for pediatric pneumonia classifi-cation on external data." Emergency Radi-ology 29.1 (2022): 107-113.
[22] Arun Prakash, J., et al. "Pediatric pneumo-nia diagnosis using stacked ensemble learn-ing on multi-model deep CNN architec-tures." Multimedia tools and applications 82.14 (2023): 21311-21351.
[23] Prakash, J. Arun, et al. "Stacked ensemble learning based on deep convolutional neural networks for pediatric pneumonia diagnosis using chest X-ray images." Neural Compu-ting and Applications 35.11 (2023): 8259-8279.
[24] Bal, Ufuk, et al. "A deep learning feature extraction-based hybrid approach for de-tecting pediatric pneumonia in chest X-ray images." Physical and Engineering Sciences in Medicine 47.1 (2024): 109-117.
[25] Tian, Y., Ye, Q., & Doermann, D. (2025). Yolov12: Attention-centric real-time object detectors. arXiv preprint arXiv:2502.12524.
[26] Zhu, Jiachen, et al. "Transformers without normalization." Proceedings of the Com-puter Vision and Pattern Recognition Con-ference. 2025.
[27] Liu, W., Lu, H., Fu, H., & Cao, Z. (2023). Learning to upsample by learning to sam-ple. In Proceedings of the IEEE/CVF inter-national conference on computer vision (pp. 6027-6037).
[28] Kermany, Daniel S., et al. "Identifying medical diagnoses and treatable diseases by image-based deep learning." cell 172.5 (2018): 1122-1131.
[29] Ming Chen, Yixuan Xu, Wanxiang Qin, Yan Li, Jiyang Yu; Tomato ripeness detection method based on FasterNet block and atten-tion mechanism. AIP Advances 1 June 2025; 15 (6): 065117. https://doi.org/10.1063/5.0280801
This work is licensed under a Creative Commons Attribution 4.0 International License.
Copyright (c) 2025 Frontiers in Preventive Medicine