Abstract

Automated sputum cell detection supports quantitative assessment of respiratory inflammation, but clinical images contain overlapping targets, heterogeneous backgrounds, large scale variation, and imbalanced classes. This study presents GRA-InternRTM, a multiclass detector based on InternImage-RTMDet. An Edge-Guided Deformable Adapter (EGDA) strengthens boundary representations, and an Adaptive Gated Fusion Module (AGFM) provides sample-specific multiscale fusion. Dynamic Head (DyHead) aligns features across scales, while Dual-Robust Learning loss (DRL loss) reduces the effects of class imbalance and missing annotations. Ablation experiments evaluated the four components on an in-house dataset. Compared with six reference detectors, GRA-InternRTM achieved mAP, mAP50, Recall, and AccMatched values of 0.365, 0.622, 0.8762, and 0.9617, respectively. Relative to Fast R-CNN, it improved mAP, mAP50, and AccMatched by 0.014, 0.024, and 0.0308. These results support the technical feasibility of multiclass detection in complex sputum cell images, although external validation remains necessary.

References

• State Council of the People's Republic of China, “Opinions of the State Council on Deepening the Implementation of the ‘Artificial Intelligence Plus’ Initiative,” Gazette of the State Council of the People's Republic of China, no. 25 (2025): 15-19, [in Chinese],
• N. Panakkal, A. Lekshmi, V. V. Saraswathy, et al., “Effective Lung Cancer Control: An Unaccomplished Challenge in Cancer Research,” CytoJournal 20 (2023): 16, doi: 10.25259/Cytojournal_36_2022.
• R. H. Pan, J. Yin, Y. Y. Wu, et al., “Association of Sputum Viscosity With Disease Severity and Prognosis in Adults With Mycoplasma pneumoniae Pneumonia,” Infectious Disease Information 39, no. 1 (2026): 80-85, [in Chinese], doi: 10.3969/j.issn.1007-8134.2026.01.013.
• L. Qin, P. G. Gibson, J. L. Simpson, et al., “Dysregulation of Sputum Columnar Epithelial Cells and Products in Distinct Asthma Phenotypes,” Clinical & Experimental Allergy 49, no. 11 (2019): 1418-1428, doi: 10.1111/cea.13452.
• K. S. Mithra and W. R. Sam Emmanuel, “Automated Identification of Mycobacterium Bacillus From Sputum Images for Tuberculosis Diagnosis,” Signal, Image and Video Processing 13, no. 8 (2019): 1585-1592, doi: 10.1007/s11760-019-01509-1.
• K. S. Mithra, “Enhanced Fuzzy Gaussian Networks for Sputum Image Based Mycobacterium Detection,” The Imaging Science Journal 71, no. 4 (2023): 313-322, doi: 10.1080/13682199.2023.2167553.
• R. Khutlang, S. Krishnan, A. Whitelaw, et al., “Automated Detection of Tuberculosis in Ziehl-Neelsen-Stained Sputum Smears Using Two One-Class Classifiers,” Journal of Microscopy 237, no. 1 (2010): 96-102, doi: 10.1111/j.1365-2818.2009.03308.x.
• Z. Jan, M. Rafiq, H. Muhammad, et al., “Detection of Tuberculosis Bacteria in Sputum Slide Image Using Morphological Features,” in Beyond Databases, Architectures and Structures (Cham, Switzerland: Springer, 2015), 408-414, doi: 10.1007/978-3-319-18422-7_36.
• Z. Le, Y. Liang, X. Hu, T. Qiu, and P. Xu, “A Risk Stratification Study of Ultrasound Images of Thyroid Nodules Based on Improved DETR,” International Journal of Imaging Systems and Technology 34, no. 6 (2024): e23219, doi: 10.1002/ima.23219.
• M. Salvi, N. Michielli, K. M. Meiburger, et al., “cyto-Knet: An Instance Segmentation Approach for Multiple Myeloma Plasma Cells Using Conditional Kernels,” International Journal of Imaging Systems and Technology 34, no. 1 (2024): e22984, doi: 10.1002/ima.22984.
• V. Biradar and K. C. Gull, “Multiple Object Detection and Tracking Using Deep Learning Framework with Non-Maximum Suppression,” International Journal of Intelligent Engineering and Systems 18, no. 1 (2025): 106-117, doi: 10.22266/ijies2025.0229.09.
• W. Wang, J. Dai, Z. Chen, et al., “InternImage: Exploring Large-Scale Vision Foundation Models With Deformable Convolutions,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, BC, Canada (Piscataway, NJ: IEEE, 2023), 14408-14419, doi: 10.1109/CVPR52729.2023.01385.
• S. Liu, L. Qi, H. Qin, et al., “Path Aggregation Network for Instance Segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA (Piscataway, NJ: IEEE, 2018), 8759-8768, doi: 10.1109/CVPR.2018.00913.
• C. Lyu, W. Zhang, H. Huang, et al., “RTMDet: An Empirical Study of Designing Real-Time Object Detectors,” arXiv preprint arXiv:2212.07784 (2022), doi: 10.48550/arXiv.2212.07784.
• R. C. Gonzalez and R. E. Woods, Digital Image Processing, 4th ed. (New York, NY: Pearson, 2018), ISBN 978-0-13-335672-4.
• T. Takikawa, D. Acuna, V. Jampani, et al., “Gated-SCNN: Gated Shape CNNs for Semantic Segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, South Korea (Piscataway, NJ: IEEE, 2019), 5228-5237, doi: 10.1109/ICCV.2019.00533.
• Z. Zhang, X. Zhang, C. Peng, et al., “ExFuse: Enhancing Feature Fusion for Semantic Segmentation,” in Computer Vision - ECCV 2018, Munich, Germany (Cham, Switzerland: Springer, 2018), 273-288, doi: 10.1007/978-3-030-01249-6_17.
• X. Li, W. Wang, X. Hu, et al., “Selective Kernel Networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA (Piscataway, NJ: IEEE, 2019), 510-519, doi: 10.1109/CVPR.2019.00060.
• X. Dai, Y. Chen, B. Xiao, et al., “Dynamic Head: Unifying Object Detection Heads With Attentions,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, TN, USA (Piscataway, NJ: IEEE, 2021), 7369-7378, doi: 10.1109/CVPR46437.2021.00729.
• J. Belda, R. Leigh, K. Parameswaran, et al., “Induced Sputum Cell Counts in Healthy Adults,” American Journal of Respiratory and Critical Care Medicine 161, no. 2 (2000): 475-478, doi: 10.1164/ajrccm.161.2.9903097.
• T.-Y. Lin, P. Goyal, R. Girshick, et al., “Focal Loss for Dense Object Detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence 42, no. 2 (2020): 318-327, doi: 10.1109/TPAMI.2018.2858826.
• Z. Zhang and M. R. Sabuncu, “Generalized Cross Entropy Loss for Training Deep Neural Networks With Noisy Labels,” in Advances in Neural Information Processing Systems 31, Montreal, Canada (Red Hook, NY: Curran Associates, 2018), 8778-8788,
• R. Girshick, “Fast R-CNN,” in Proceedings of the IEEE International Conference on Computer Vision, Santiago, Chile (Piscataway, NJ: IEEE, 2015), 1440-1448, doi: 10.1109/ICCV.2015.169.
• Z. Cai and N. Vasconcelos, “Cascade R-CNN: Delving Into High Quality Object Detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA (Piscataway, NJ: IEEE, 2018), 6154-6162, doi: 10.1109/CVPR.2018.00644.
• K. He, G. Gkioxari, P. Dollar, et al., “Mask R-CNN,” IEEE Transactions on Pattern Analysis and Machine Intelligence 42, no. 2 (2020): 386-397, doi: 10.1109/TPAMI.2018.2844175.
• G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics YOLOv8,” version 8.0.0 (2023),
• A. Wang, H. Chen, L. Liu, et al., “YOLOv10: Real-Time End-to-End Object Detection,” Advances in Neural Information Processing Systems 37 (2024): 107984-108011, doi: 10.52202/079017-3429.
• Z. Liu, Y. Lin, Y. Cao, et al., “Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, QC, Canada (Piscataway, NJ: IEEE, 2021), 9992-10002, doi: 10.1109/ICCV48922.2021.00986.