Home Health Care Science Article
PDF (3.4 MB)
Cite
EndNote(RIS) BibTeX
Collect
Submit Manuscript
Show Outline
Figures (4)

Tables (8)
TABLE 1
TABLE 2
TABLE 3
TABLE 4
TABLE 5
Show 3 more tables Hide 3 tables
Original Article | Open Access

Leveraging anatomical constraints with uncertainty for pneumothorax segmentation

Han Yuan1, Chuan Hong2, Nguyen Tuan Anh Tran3, Xinxing Xu4, Nan Liu1,5,6 ()
Centre for Quantitative Medicine, Duke‐NUS Medical School, Singapore
Department of Biostatistics and Bioinformatics, Duke University, Durham, North Carolina, USA
Department of Diagnostic Radiology, Singapore General Hospital, Singapore
Institute of High Performance Computing, Agency for Science, Technology and Research, Singapore
Programme in Health Services and Systems Research, Duke‐NUS Medical School, Singapore
Institute of Data Science, National University of Singapore, Singapore

Han Yuan and Chuan Hong contributed equally to this study.

Show Author Information

Graphical Abstract

View original image Download original image
We proposed a novel approach that incorporates the lung + space as a constraint during deep learning model training for pneumothorax segmentation on 2D chest radiographs. We utilized external datasets and an auxiliary task of lung segmentation to generate a specific constraint of lung + space for each chest radiograph, circumventing the need for additional annotations. We incorporated a discriminator to eliminate unreliable constraints caused by the domain shift between the auxiliary and target datasets. Our results demonstrated consistent improvements across six baseline models built on three segmentation architectures and two convolutional neural networks backbones.

Abstract

Background

Pneumothorax is a medical emergency caused by the abnormal accumulation of air in the pleural space—the potential space between the lungs and chest wall. On 2D chest radiographs, pneumothorax occurs within the thoracic cavity and outside of the mediastinum, and we refer to this area as “lung + space.” While deep learning (DL) has increasingly been utilized to segment pneumothorax lesions in chest radiographs, many existing DL models employ an end‐to‐end approach. These models directly map chest radiographs to clinician‐annotated lesion areas, often neglecting the vital domain knowledge that pneumothorax is inherently location‐sensitive.

Methods

We propose a novel approach that incorporates the lung + space as a constraint during DL model training for pneumothorax segmentation on 2D chest radiographs. To circumvent the need for additional annotations and to prevent potential label leakage on the target task, our method utilizes external datasets and an auxiliary task of lung segmentation. This approach generates a specific constraint of lung + space for each chest radiograph. Furthermore, we have incorporated a discriminator to eliminate unreliable constraints caused by the domain shift between the auxiliary and target datasets.

Results

Our results demonstrated considerable improvements, with average performance gains of 4.6%, 3.6%, and 3.3% regarding intersection over union, dice similarity coefficient, and Hausdorff distance. These results were consistent across six baseline models built on three architectures (U‐Net, LinkNet, or PSPNet) and two backbones (VGG‐11 or MobileOne‐S0). We further conducted an ablation study to evaluate the contribution of each component in the proposed method and undertook several robustness studies on hyper‐parameter selection to validate the stability of our method.

Conclusions

The integration of domain knowledge in DL models for medical applications has often been underemphasized. Our research underscores the significance of incorporating medical domain knowledge about the location‐specific nature of pneumothorax to enhance DL‐based lesion segmentation and further bolster clinicians' trust in DL tools. Beyond pneumothorax, our approach is promising for other thoracic conditions that possess location‐relevant characteristics.

Keywords

constrained optimizationdeep transfer learningdiagnostic radiologypneumothorax detectionsemantic segmentation

References

1

Imran JB, Eastman AL. Pneumothorax. JAMA. 2017;318(10):974. https://doi.org/10.1001/jama.2017.10476

Crossref Google Scholar
2

Sahn SA, Heffner JE. Spontaneous pneumothorax. N Engl J Med. 2000;342(12):868–74. https://doi.org/10.1056/NEJM200003233421207

Crossref Google Scholar
3

Çallı E, Sogancioglu E, van Ginneken B, van Leeuwen KG, Murphy K. Deep learning for chest X‐ray analysis: a survey. Med Image Anal. 2021;72:102125. https://doi.org/10.1016/j.media.2021.102125

Crossref Google Scholar
4

Noppen M, De Keukeleire T. Pneumothorax. Respiration. 2008;76(2):121–7. https://doi.org/10.1159/000135932

Crossref Google Scholar
5

Ding W, Shen Y, Yang J, He X, Zhang M. Diagnosis of pneumothorax by radiography and ultrasonography. Chest. 2011;140(4):859–66. https://doi.org/10.1378/chest.10-2946

Crossref Google Scholar
6

Gu D, Su K, Zhao H. A case‐based ensemble learning system for explainable breast cancer recurrence prediction. Artif Intell Med. 2020;107:101858. https://doi.org/10.1016/j.artmed.2020.101858

Crossref Google Scholar
7

Hong W, Hwang EJ, Lee JH, Park J, Goo JM, Park CM. Deep learning for detecting pneumothorax on chest radiographs after needle biopsy: clinical implementation. Radiology. 2022;303(2):433–41. https://doi.org/10.1148/radiol.211706

Crossref Google Scholar
8

Sharma N, Aggarwal L. Automated medical image segmentation techniques. J Med Phys. 2010;35(1):3–14. https://doi.org/10.4103/0971-6203.58777

Crossref Google Scholar
9

Duncan JS, Ayache N. Medical image analysis: progress over two decades and the challenges ahead. IEEE Trans Pattern Anal Mach Intell. 2000;22(1):85–106. https://doi.org/10.1109/34.824822

Crossref Google Scholar
10

McInerney T, Terzopoulos D. Deformable models in medical image analysis: a survey. Med Image Anal. 1996;1(2):91–108. https://doi.org/10.1016/s1361-8415(96)80007-7

Crossref Google Scholar
11

Pham DL, Xu C, Prince JL. Current methods in medical image segmentation. Annu Rev Biomed Eng. 2000;2:315–37. https://doi.org/10.1146/annurev.bioeng.2.1.315

Crossref Google Scholar
12

Grau V, Mewes AUJ, Alcaniz M, Kikinis R, Warfield SK. Improved watershed transform for medical image segmentation using prior information. IEEE Trans Med Imaging. 2004;23(4):447–58. https://doi.org/10.1109/TMI.2004.824224

Crossref Google Scholar
13

Ahmed MN, Yamany SM, Mohamed N, Farag AA, Moriarty T. A modified fuzzy C‐means algorithm for bias field estimation and segmentation of MRI data. IEEE Trans Med Imaging. 2002;21(3):193–9. https://doi.org/10.1109/42.996338

Crossref Google Scholar
14

Xie F, Yuan H, Ning Y, Ong MEH, Feng M, Hsu W, et al. Deep learning for temporal data representation in electronic health records: a systematic review of challenges and methodologies. J Biomed Inf. 2022;126:103980. https://doi.org/10.1016/j.jbi.2021.103980

Crossref Google Scholar
15

Wang Q, Liu Q, Luo G, Liu Z, Huang J, Zhou Y, et al. Automated segmentation and diagnosis of pneumothorax on chest X‐rays with fully convolutional multi‐scale ScSE‐DenseNet: a retrospective study. BMC Med Inform Decis Mak. 2020;20(Suppl 14):317. https://doi.org/10.1186/s12911-020-01325-5

Crossref Google Scholar
16

Iqbal T, Shaukat A, Akram MU, Mustansar Z, Khan A. Automatic diagnosis of pneumothorax from chest radiographs: a systematic literature review. IEEE Access. 2021;9:145817–39. https://doi.org/10.1109/ACCESS.2021.3122998

Crossref Google Scholar
17

McBee MP, Awan OA, Colucci AT, Ghobadi CW, Kadom N, Kansagra AP, et al. Deep learning in radiology. Academic Radiol. 2018;25(11):1472–80. https://doi.org/10.1016/j.acra.2018.02.018

Crossref Google Scholar
18

Moses DA. Deep learning applied to automatic disease detection using chest X‐rays. J Med Imaging Radiat Oncol. 2021;65(5):498–517. https://doi.org/10.1111/1754-9485.13273

Crossref Google Scholar
19

Tang YX, Tang YB, Peng Y, Yan K, Bagheri M, Redd BA, et al. Automated abnormality classification of chest radiographs using deep convolutional neural networks. npj Digital Med. 2020;3:70. https://doi.org/10.1038/s41746-020-0273-z

Crossref Google Scholar
20

Krizhevsky A, Sutskever I, Hinton GE. ImageNet classification with deep convolutional neural networks. Commun ACM. 2017;60(6):84–90. https://doi.org/10.1145/3065386

Crossref Google Scholar
21
Simonyan K, Zisserman A. Very deep convolutional networks for large‐scale image recognition. 3rd Int Conf Learn Represent ICLR 2015 Conf Track Proc. 2015:1–14. https://arxiv.org/abs/1409.1556
22
Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, et al. Going deeper with convolutions. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA, 2015:1–9. https://doi.org/10.1109/CVPR.2015.7298594
Crossref
23

Thian YL, Ng D, Hallinan JTPD, Jagmohan P, Sia SY, Tan CH, et al. Deep learning systems for pneumothorax detection on chest radiographs: a multicenter external validation study. Radiol: Artif Intell. 2021;3(4):e200190. https://doi.org/10.1148/ryai.2021200190

Crossref Google Scholar
24
Tan M, Le QV EfficientNet: rethinking model scaling for convolutional neural networks. 2019. arXiv. http://arxiv.org/abs/1905.11946
25

Wang Y, Sun L, Jin Q. Enhanced diagnosis of pneumothorax with an improved real‐time augmentation for imbalanced chest X‐rays data based on DCNN. IEEE/ACM Trans Comput Biol Bioinf. 2021;18(3):951–62. https://doi.org/10.1109/TCBB.2019.2911947

Crossref Google Scholar
26

Feng S, Liu Q, Patel A, Bazai SU, Jin CK, Kim JS, et al. Automated pneumothorax triaging in chest X‐rays in the New Zealand population using deep‐learning algorithms. J Med Imaging Radiat Oncol. 2022;66(8):1035–43. https://doi.org/10.1111/1754-9485.13393

Crossref Google Scholar
27
He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA. 2016:770–8. https://doi.org/10.1109/CVPR.2016.90
Crossref
28
Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 2017:2261–9. https://doi.org/10.1109/CVPR.2017.243
Crossref
29

Mosquera C, Diaz FN, Binder F, Rabellino JM, Benitez SE, Beresñak AD, et al. Chest X‐ray automated triage: a semiologic approach designed for clinical implementation, exploiting different types of labels through a combination of four deep learning architectures. Comput Methods Programs Biomed. 2021;206:106130. https://doi.org/10.1016/j.cmpb.2021.106130

Crossref Google Scholar
30
Shvets AA, Rakhlin A, Kalinin AA, Iglovikov VI. Automatic Instrument Segmentation in Robot‐Assisted Surgery using Deep Learning. 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), Orlando, FL, USA, 2018:624–8. https://doi.org/10.1109/ICMLA.2018.00100
Crossref
31
Rocha J, Pereira SC, Pedrosa J, Campilho A, Mendonça AM. Attention‐driven spatial transformer network for abnormality detection in chest X‐ray images. 2022 IEEE 35th International Symposium on Computer‐Based Medical Systems (CBMS), Shenzhen, China, 2022:252–7. https://doi.org/10.1109/CBMS55023.2022.00051
Crossref
32
Jaderberg MSimonyan KZisserman ASpatial transformer networks. Montreal, Canada: Advances in Neural Information Processing Systems (NeurIPS); 2015. https://arxiv.org/abs/1506.02025
33

Chen KC, Yu HR, Chen WS, Lin WC, Lee YC, Chen HH, et al. Diagnosis of common pulmonary diseases in children by X‐ray images and deep learning. Sci Rep. 2020;10(1):17374. https://doi.org/10.1038/s41598-020-73831-5

Crossref Google Scholar
34
Redmon J, Divvala S, Girshick R, Farhadi A. You only look once: unified, real‐time object detection. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016:779–88. https://doi.org/10.1109/CVPR.2016.91
Crossref
35

Li Y, Wong D, Sreng S, Chung J, Toh A, Yuan H, et al. Effect of childhood atropine treatment on adult choroidal thickness using sequential deep learning‐enabled segmentation. Asia‐Pacific J Ophthalmol. 2024;13(5):100107. https://doi.org/10.1016/j.apjo.2024.100107

Crossref Google Scholar
36

Saporta A, Gui X, Agrawal A, Pareek A, Truong SQH, Nguyen CDT, et al. Benchmarking saliency methods for chest X‐ray interpretation. Nat Mach Intell. 2022;4:867–78. https://doi.org/10.1038/s42256-022-00536-x

Crossref Google Scholar
37

Agrawal T, Choudhary P. Segmentation and classification on chest radiography: a systematic survey. Visual Comput. 2023;39(3):875–913. https://doi.org/10.1007/s00371-021-02352-7

Crossref Google Scholar
38
Ronneberger OFischer PBrox TU‐net: convolutional networks for biomedical image segmentation. In: Lecture notes in computer science. Cham: Springer International Publishing; 2015. p. 234–241. https://doi.org/10.1007/978-3-319-24574-4_28
Crossref
39
Abedalla A, Abdullah M, Al‐Ayyoub M, Benkhelifa E. 2ST‐UNet: 2‐Stage Training Model using U‐Net for pneumothorax segmentation in chest X‐rays. 2020 International Joint Conference on Neural Networks (IJCNN), Glasgow, UK, 2020:1–6. https://doi.org/10.1109/IJCNN48605.2020.9207268
Crossref
40

Wang X, Yang S, Lan J, Fang Y, He J, Wang M, et al. Automatic segmentation of pneumothorax in chest radiographs based on a two‐stage deep learning method. IEEE Trans Cogn Dev Syst. 2022;14(1):205–18. https://doi.org/10.1109/TCDS.2020.3035572

Crossref Google Scholar
41

Chen LC, Papandreou G, Kokkinos I, Murphy K, Yuille AL. DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. IEEE Trans Pattern Anal Mach Intell. 2018;40(4):834–48. https://doi.org/10.1109/TPAMI.2017.2699184

Crossref Google Scholar
42
He K, Gkioxari G, Dollár P, Girshick R, Mask R‐CNN. 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 2017:2980–8.
Crossref
43

Tolkachev A, Sirazitdinov I, Kholiavchenko M, Mustafaev T, Ibragimov B. Deep learning for diagnosis and segmentation of pneumothorax: the results on the kaggle competition and validation against radiologists. IEEE J Biomed Health Inform. 2021;25(5):1660–72. https://doi.org/10.1109/JBHI.2020.3023476

Crossref Google Scholar
44

Lee SY, Ha S, Jeon MG, Li H, Choi H, Kim HP, et al. Localization‐adjusted diagnostic performance and assistance effect of a computer‐aided detection system for pneumothorax and consolidation. npj Digital Med. 2022;5(1):107. https://doi.org/10.1038/s41746-022-00658-x

Crossref Google Scholar
45

Yuan H, Yu K, Xie F, Liu M, Sun S. Automated machine learning with interpretation: a systematic review of methodologies and applications in healthcare. Med Adv. 2024;2(3):205–37. https://doi.org/10.1002/med4.75

Crossref Google Scholar
46

Pereira S, Meier R, McKinley R, Wiest R, Alves V, Silva CA, et al. Enhancing interpretability of automatically extracted machine learning features: application to a RBM‐Random Forest system on brain lesion segmentation. Med Image Anal. 2018;44:228–44. https://doi.org/10.1016/j.media.2017.12.009

Crossref Google Scholar
47

Yuan H, Kang L, Li Y, Fan Z. Human‐in‐the‐loop machine learning for healthcare: current progress and future opportunities in electronic health records. Med Adv. 2024;2(3):318–22. https://doi.org/10.1002/med4.70

Crossref Google Scholar
48

Van Ginneken B, Ter Haar Romeny BM, Viergever MA. Computer‐aided diagnosis in chest radiography: a survey. IEEE Trans Med Imaging. 2001;20(12):1228–41. https://doi.org/10.1109/42.974918

Crossref Google Scholar
49

Li F, Armato SG, Engelmann R, Rhines T, Crosby J, Lan L, et al. Anatomic point‐based lung region with zone identification for radiologist annotation and machine learning for chest radiographs. J Digit Imaging. 2021;34(4):922–31. https://doi.org/10.1007/s10278-021-00494-7

Crossref Google Scholar
50
Crosby J, Rhines T, Li F, MacMahon H, Giger M. Deep learning for pneumothorax detection and localization using networks fine‐tuned with multiple institutional datasetsMedical Imaging 2020: Computer‐Aided Diagnosis, Houston, USA, 2020. https://doi.org/10.1117/12.2549709
Crossref
51

Jung HG, Nam WJ, Kim HW, Lee SW. Weakly supervised thoracic disease localization via disease masks. Neurocomputing. 2023;517:34–43. https://doi.org/10.1016/j.neucom.2022.10.019

Crossref Google Scholar
52

Bateson M, Dolz J, Kervadec H, Lombaert H, Ayed IB. Constrained domain adaptation for image segmentation. IEEE Trans Med Imaging. 2021;40(7):1875–87. https://doi.org/10.1109/TMI.2021.3067688

Crossref Google Scholar
53
Bateson M, Kervadec H, Dolz J, Lombaert H, Ben Ayed I. Constrained domain adaptation for segmentation. 2019 International conference on medical image computing and computer‐assisted intervention (MICCAI), Shenzhen, China, 2019:326‐334. https://doi.org/10.1007/978-3-030-32245-8_37
Crossref
54

Kervadec H, Dolz J, Tang M, Granger E, Boykov Y, Ben Ayed I. Constrained‐CNN losses for weakly supervised segmentation. Med Image Anal. 2019;54:88–99. https://doi.org/10.1016/j.media.2019.02.009

Crossref Google Scholar
55

van Ginneken B, Katsuragawa S, ter Haar Romeny BM, Kunio Doi D, Viergever MA. Automatic detection of abnormalities in chest radiographs using local texture analysis. IEEE Trans Med Imaging. 2002;21(2):139–49. https://doi.org/10.1109/42.993132

Crossref Google Scholar
56

Shiraishi J, Katsuragawa S, Ikezoe J, Matsumoto T, Kobayashi T, Komatsu K, et al. Development of a digital image database for chest radiographs with and without a lung nodule: receiver operating characteristic analysis of radiologists' detection of pulmonary nodules. AJR Am J Roentgenol. 2000;174(1):71–4. https://doi.org/10.2214/ajr.174.1.1740071

Crossref Google Scholar
57

Jaeger S, Candemir S, Antani S, Wáng YX, Lu PX, Thoma G. Two public chest X‐ray datasets for computer‐aided screening of pulmonary diseases. Quant Imaging Med Surg. 2014;4(6):475–7. https://doi.org/10.3978/j.issn.2223-4292.2014.11.20

Crossref Google Scholar
58
Bateson MKervadec HDolz JLombaert HBen Ayed ISource‐relaxed domain adaptation for image segmentation. In: Lecture notes in computer science. Cham: Springer International Publishing; 2020. p. 490–499. https://doi.org/10.1007/978-3-030-59710-8_48
Crossref
59

Huang X, Yang X, Dou H, Huang Y, Zhang L, Liu Z, et al. Test‐time bi‐directional adaptation between image and model for robust segmentation. Comput Methods Programs Biomed. 2023;233:107477. https://doi.org/10.1016/j.cmpb.2023.107477

Crossref Google Scholar
60

Larrazabal AJ, Martinez C, Glocker B, Ferrante E. Post‐DAE: anatomically plausible segmentation via post‐processing with denoising autoencoders. IEEE Trans Med Imaging. 2020;39(12):3813–20. https://doi.org/10.1109/TMI.2020.3005297

Crossref Google Scholar
61

Souza JC, Bandeira Diniz JO, Ferreira JL, França da Silva GL, Corrêa Silva A, de Paiva AC. An automatic method for lung segmentation and reconstruction in chest X‐ray using deep neural networks. Comput Methods Programs Biomed. 2019;177:285–96. https://doi.org/10.1016/j.cmpb.2019.06.005

Crossref Google Scholar
62

Candemir S, Jaeger S, Palaniappan K, Musco JP, Singh RK, Zhiyun Xue X, et al. Lung segmentation in chest radiographs using anatomical atlases with nonrigid registration. IEEE Trans Med Imaging. 2014;33(2):577–90. https://doi.org/10.1109/TMI.2013.2290491

Crossref Google Scholar
63

Mansoor A, Bagci U, Xu Z, Foster B, Olivier KN, Elinoff JM, et al. A generic approach to pathological lung segmentation. IEEE Trans Med Imaging. 2014;33(12):2293–310. https://doi.org/10.1109/TMI.2014.2337057

Crossref Google Scholar
64

Mansoor A, Cerrolaza JJ, Perez G, Biggs E, Okada K, Nino G, et al. A generic approach to lung field segmentation from chest radiographs using deep space and shape learning. IEEE Trans Biomed Eng. 2020;67(4):1206–20. https://doi.org/10.1109/TBME.2019.2933508

Crossref Google Scholar
65
Yuan H, Jiang P, Zhao G. Human‐guided design to explain deep learning‐based pneumothorax classifier. 2023 International Conference on Medical Imaging with Deep Learning (MIDL), Nashville, TN, USA, 2023:1–4. https://openreview.net/forum?id=_kk8KI8MiRE
66

Yuan H, Hong C, Jiang PT, Zhao G, Tran NTA, Xu X, et al. Clinical domain knowledge‐derived template improves post hoc AI explanations in pneumothorax classification. J Biomed Inf. 2024;156:104673. https://doi.org/10.1016/j.jbi.2024.104673

Crossref Google Scholar
67

Bateson M, Kervadec H, Dolz J, Lombaert H, Ben Ayed I. Source‐free domain adaptation for image segmentation. Med Image Anal. 2022;82:102617. https://doi.org/10.1016/j.media.2022.102617

Crossref Google Scholar
68

Frank O, Schipper N, Vaturi M, Soldati G, Smargiassi A, Inchingolo R, et al. Integrating domain knowledge into deep networks for lung ultrasound with applications to COVID‐19. IEEE Trans Med Imaging. 2022;41(3):571–81. https://doi.org/10.1109/TMI.2021.3117246

Crossref Google Scholar
69

Ma J, Chen J, Ng M, Huang R, Li Y, Li C, et al. Loss odyssey in medical image segmentation. Med Image Anal. 2021;71:102035. https://doi.org/10.1016/j.media.2021.102035

Crossref Google Scholar
70
Klodt M, Cremers D. A convex framework for image segmentation with moment constraints. 2011 International Conference on Computer Vision, Barcelona, Spain, 2011:2236–2243. https://doi.org/10.1109/ICCV.2011.6126502
Crossref
71

Wang L, Wang L, Chen KC, Shi F, Liao S, Li G, et al. Automated segmentation of CBCT image using spiral CT atlases and convex optimization. Med Image Comput Comput Assist Interv. 2013;16(pt 3):251–8. https://doi.org/10.1007/978-3-642-40760-4_32

Crossref Google Scholar
72
Langer SG, Shih G. SIIM‐ACR Pneumothorax Segmentation Dataset. 2019. https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation
73

van Ginneken B, Stegmann MB, Loog M. Segmentation of anatomical structures in chest radiographs using supervised methods: a comparative study on a public database. Med Image Anal. 2006;10(1):19–40. https://doi.org/10.1016/j.media.2005.02.002

Crossref Google Scholar
74
Vasu PKA, Gabriel J, Zhu J, Tuzel O, Ranjan A. MobileOne: an improved one millisecond mobile backbone. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada, 2023:7907–7917. https://doi.org/10.1109/CVPR52729.2023.00764
Crossref
75
Wang X, Peng Y, Lu L, Lu Z, Bagheri M, Summers RM. ChestX‐Ray8:hospital‐scale chest X‐ray database and benchmarks on weakly‐supervised classification and localization of common Thorax diseases. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 2017:3462–71. https://doi.org/10.1109/CVPR.2017.369
Crossref
76

Gut D, Tabor Z, Szymkowski M, Rozynek M, Kucybala I, Wojciechowski W. Benchmarking of deep architectures for segmentation of medical images. IEEE Trans Med Imaging. 2022;41(11):3231–41. https://doi.org/10.1109/TMI.2022.3180435

Crossref Google Scholar
77
Chaurasia A, Culurciello E. LinkNet: exploiting encoder representations for efficient semantic segmentation. 2017 IEEE Visual Communications and Image Processing (VCIP), St. Petersburg, FL, USA, 2017:1–4. https://doi.org/10.1109/VCIP.2017.8305148
Crossref
78
Zhao H, Shi J, Qi X, Wang X, Jia J. Pyramid scene parsing network. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 2017:6230–6239. https://doi.org/10.1109/CVPR.2017.660
Crossref
79

Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier‐Hein KH. nnU‐Net: a self‐configuring method for deep learning‐based biomedical image segmentation. Nat Methods. 2021;18(2):203–11. https://doi.org/10.1038/s41592-020-01008-z

Crossref Google Scholar
80
Kirillov A, Mintun E, Ravi N, Mao H, Rolland C, Gustafson L, et al. Segment anything. 2023 IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, 2023:3992–4003. https://doi.org/10.1109/ICCV51070.2023.00371
Crossref
81

Cai Z, Vasconcelos N. Cascade R‐CNN: high quality object detection and instance segmentation. IEEE Trans Pattern Anal Mach Intell. 2021;43(5):1483–98. https://doi.org/10.1109/TPAMI.2019.2956516

Crossref Google Scholar
82
Yuan H. Constrained learning code. 2023. https://github.com/Han-Yuan-Med/constrained-learning
83

Foo LL, Lim GYS, Lanca C, Wong CW, Hoang QV, Zhang XJ, et al. Deep learning system to predict the 5‐year risk of high myopia using fundus imaging in children. npj Digital Med. 2023;6(1):10. https://doi.org/10.1038/s41746-023-00752-8

Crossref Google Scholar
84

Jaccard P. The distribution of the flora in the alpine zone.1. New Phytol. 1912;11:37–50. https://doi.org/10.1111/J.1469-8137.1912.TB05611.X

Crossref Google Scholar
85

Dice LR. Measures of the amount of ecologic association between species. Ecology. 1945;26(3):297–302. https://doi.org/10.2307/1932409

Crossref Google Scholar
86

Beauchemin M, Thomson KPB, Edwards G. On the Hausdorff distance used for the evaluation of segmentation results. Can J Remote Sensing. 1998;24(1):3–8. https://doi.org/10.1080/07038992.1998.10874685

Crossref Google Scholar
87

Umesh Adiga PS, Chaudhuri BB. An efficient method based on watershed and rule‐based merging for segmentation of 3‐D histo‐pathological images. Pattern Recogn. 2001;34(7):1449–58. https://doi.org/10.1016/s0031-3203(00)00076-5

Crossref Google Scholar
88

Wright T, Klein M, Wieczorek J. A primer on visualizations for comparing populations, including the issue of overlapping confidence intervals. Am Stat. 2019;73(2):165–78. https://doi.org/10.1080/00031305.2017.1392359

Crossref Google Scholar
89

El Jurdi R, Petitjean C, Honeine P, Cheplygina V, Abdallah F. High‐level prior‐based loss functions for medical image segmentation: a survey. Comput Vis Image Underst. 2021;210:103248. https://doi.org/10.1016/j.cviu.2021.103248

Crossref Google Scholar
90

Paul A, Tang YX, Shen TC, Summers RM. Discriminative ensemble learning for few‐shot chest X‐ray diagnosis. Med Image Anal. 2021;68:101911. https://doi.org/10.1016/j.media.2020.101911

Crossref Google Scholar
91

Mittal A, Hooda R, Sofat S. Lung field segmentation in chest radiographs: a historical review, current status, and expectations from deep learning. IET Image Processing. 2017;11(11):937–52. https://doi.org/10.1049/iet-ipr.2016.0526

Crossref Google Scholar
92

Wang B, Takeda T, Sugimoto K, Zhang J, Wada S, Konishi S, et al. Automatic creation of annotations for chest radiographs based on the positional information extracted from radiographic image reports. Comput Methods Programs Biomed. 2021;209:106331. https://doi.org/10.1016/j.cmpb.2021.106331

Crossref Google Scholar
93

Noble JA, Boukerroui D. Ultrasound image segmentation: a survey. IEEE Trans Med Imaging. 2006;25(8):987–1010. https://doi.org/10.1109/tmi.2006.877092

Crossref Google Scholar
94

Chen H, Qi X, Yu L, Dou Q, Qin J, Heng PA. DCAN: deep contour‐aware networks for object instance segmentation from histology images. Med Image Anal. 2017;36:135–46. https://doi.org/10.1016/j.media.2016.11.004

Crossref Google Scholar
95

Heimann T, Meinzer HP. Statistical shape models for 3D medical image segmentation: a review. Med Image Anal. 2009;13(4):543–63. https://doi.org/10.1016/j.media.2009.05.004

Crossref Google Scholar
96

Shamshad F, Khan S, Zamir SW, Khan MH, Hayat M, Khan FS, et al. Transformers in medical imaging: a survey. Med Image Anal. 2023;88:102802. https://doi.org/10.1016/j.media.2023.102802

Crossref Google Scholar
97

Lind Plesner L, Müller FC, Brejnebøl MW, Laustrup LC, Rasmussen F, Nielsen OW, et al. Commercially available chest radiograph AI tools for detecting airspace disease, pneumothorax, and pleural effusion. Radiology. 2023;308(3):e231236. https://doi.org/10.1148/radiol.231236

Crossref Google Scholar
98

Afrose S, Song W, Nemeroff CB, Lu C, Yao D. Subpopulation‐specific machine learning prognosis for underrepresented patients with double prioritized bias correction. Communications Medicine. 2022;2:111. https://doi.org/10.1038/s43856-022-00165-w

Crossref Google Scholar
99

Yuan H. Toward real‐world deployment of machine learning for health care: external validation, continual monitoring, and randomized clinical trials. Health Care Science. 2024;3(5):360–4. https://doi.org/10.1002/hcs2.114

Crossref Google Scholar
100

Hu Q, de F Souza LF, Holanda GB, Alves SSA, Dos S Silva FH, Han T, et al. An effective approach for CT lung segmentation using mask region‐based convolutional neural networks. Artif Intell Med. 2020;103:101792. https://doi.org/10.1016/j.artmed.2020.101792

Crossref Google Scholar
101

Azizi S, Culp L, Freyberg J, Mustafa B, Baur S, Kornblith S, et al. Robust and data‐efficient generalization of self‐supervised machine learning for diagnostic imaging. Nat Biomed Eng. 2023;7(6):756–79. https://doi.org/10.1038/s41551-023-01049-7

Crossref Google Scholar
102

Mao C, Yao L, Luo Y. ImageGCN: multi‐relational image graph convolutional networks for disease identification with chest X‐rays. IEEE Trans Med Imaging. 2022;41(8):1990–2003. https://doi.org/10.1109/TMI.2022.3153322

Crossref Google Scholar
Health Care Science
Volume 3 Issue 6,
December 2024
Pages 456-474
DOI: 10.1002/hcs2.119
Cite this article:
Yuan H, Hong C, Tran NTA, et al. Leveraging anatomical constraints with uncertainty for pneumothorax segmentation. Health Care Science, 2024, 3(6): 456-474. https://doi.org/10.1002/hcs2.119
Metrics & Citations  
Article History
Copyright
Rights and Permissions
Return

An overview of the used data set, abbreviation, and function.

Data set nameAbbreviationPurpose
Japanese Society of Radiological Technology data set [56]JSRTLung segmentationLung + space segmentation
Shenzhen data set [57]Shenzhen
Montgomery County data set [57]MC
Society for Imaging Informatics in Medicine‐American College of Radiology Pneumothorax Segmentation data set [72]SIIM‐ACRLung + space discrimination
Pneumothorax segmentation

Default experimental settings in constrained segmentation.

PhaseHyper‐parameterCandidate
Auxiliary lung segmentationArchitectureU‐Net
BackboneVGG‐11
Loss functionDice loss
OptimizerSGD
Lung + space segmentationClosing element size19 × 19
Dilation element size15 × 15
Lung + space discriminationCover rate0.99
BackboneVGG‐11
Loss functionCross‐entropy
OptimizerSGD
Pneumothorax segmentationArchitectureU‐Net, LinkNet, PSPNet
BackboneVGG‐11, MobileOne‐S0
Loss functionDice loss
OptimizerSGD

Performance of the lung segmenter on the external lung segmentation test datasets of JSRT, Shenzhen, and MC, in terms of IoU, DSC, and HD.

DatasetsIoUDSCHD
Note: The mean value of each measurement is presented alongside its corresponding confidence interval, enclosed within brackets. Higher DSC or IoU values indicate better performance, while lower HD values demonstrate better performance.Abbreviations: DSC, dice similarity coefficient; HD, Hausdorff distance; IoU, intersection over union.
JSRT0.949 (0.943–0.955)0.974 (0.972–0.976)3.814 (3.536–4.092)
Shenzhen0.918 (0.902–0.934)0.956 (0.946–0.966)4.158 (3.807–4.509)
MC0.961 (0.955–0.967)0.980 (0.976–0.984)3.732 (3.432–4.032)

Classification results of the lung + space segmentation discriminator on the test data set of the SIIM‐ACR data set.

Cutoff valueaAUROCSpecificitySensitivityPPVNPV
Note: The mean value of each metric is presented alongside its corresponding confidence interval, enclosed within brackets.Abbreviations: NPV, negative predictive value; PPV, positive predictive value.aCutoff value was incrementally determined with a step of 0.1, utilizing specificity thresholds of 0.80, 0.85, 0.90, and 0.95 on the validation set. In our experiments, as cutoff values were escalated, shifts in specificity values were observed on the validation set, resulting in identical cutoff values across distinct specificity thresholds of 0.85 and 0.90.
0.680.750 (0.705–0.795)0.854 (0.793–0.915)0.394 (0.333–0.455)0.846 (0.787–0.905)0.410 (0.353–0.467)
0.700.924 (0.875–0.973)0.244 (0.193–0.295)0.867 (0.787–0.947)0.376 (0.323–0.429)
0.700.924 (0.875–0.973)0.244 (0.193–0.295)0.867 (0.787–0.947)0.376 (0.323–0.429)
0.720.962 (0.935–0.989)0.147 (0.114–0.180)0.887 (0.813–0.961)0.358 (0.309–0.407)

Performance comparison of the constrained and baseline segmentation with different architectures and backbones, in terms of IoU, DSC, and HD.

ArchitecturesBackbonesMethodsIoUDSCHD
Note: The mean value of each measurement is presented alongside its corresponding confidence interval, enclosed within brackets. Higher DSC or IoU values indicate better performance, while lower HD values demonstrate better performance.Abbreviations: DSC, dice similarity coefficient; HD, Hausdorff distance; IoU, intersection over union.
U‐NetVGG‐11Baseline0.316 (0.296–0.336)0.441 (0.417–0.465)4.799 (4.681–4.917)
Ours0.336 (0.316–0.356)0.461 (0.437–0.485)4.558 (4.454–4.662)
Improvement6.3%4.5%5.0%
MobileOne‐S0Baseline0.309 (0.287–0.331)0.431 (0.404–0.458)4.703 (4.605–4.801)
Ours0.326 (0.306–0.346)0.449 (0.427–0.471)4.586 (4.496–4.676)
Improvement5.5%4.2%2.5%
LinkNetVGG‐11Baseline0.305 (0.287–0.323)0.426 (0.404–0.448)4.740 (4.652–4.828)
Ours0.322 (0.300–0.344)0.447 (0.422–0.472)4.592 (4.490–4.694)
Improvement5.6%4.9%3.1%
MobileOne‐S0Baseline0.302 (0.284–0.320)0.425 (0.403–0.447)4.839 (4.743–4.935)
Ours0.320 (0.300–0.340)0.447 (0.423–0.471)4.675 (4.589–4.761)
Improvement6.0%5.2%3.4%
PSPNetVGG‐11Baseline0.302 (0.282–0.322)0.424 (0.400–0.448)4.866 (4.768–4.964)
Ours0.307 (0.289–0.325)0.429 (0.407–0.451)4.660 (4.558–4.762)
Improvement1.7%1.2%4.2%
MobileOne‐S0Baseline0.260 (0.242–0.278)0.377 (0.355–0.399)5.008 (4.900–5.116)
Ours0.267 (0.247–0.287)0.382 (0.358–0.406)4.935 (4.831–5.039)
Improvement2.7%1.3%1.5%
Average improvement (%)4.63.63.3

Ablation study on the lung + space segmentation model and the lung + space segmentation discriminator.

MethodsLung area segmenterLung + space segmenterLung + space discriminatorIoUDSCHD
Abbreviations: DSC, dice similarity coefficient; HD, Hausdorff distance; IoU, intersection over union.
Baseline×××0.316 (0.296–0.336)0.441 (0.417–0.465)4.799 (4.681–4.917)
√××0.298 (0.278–0.318)0.421 (0.397–0.445)4.895 (4.791–4.999)
×√×0.317 (0.297–0.337)0.439 (0.415–0.463)4.770 (4.682–4.858)
Ours×√√0.336 (0.316–0.356)0.461 (0.437–0.485)4.558 (4.454–4.662)

Robustness study on morphological element sizes in the lung + space segmenter.

MethodsClosing element sizeDilation element sizeIoUDSCHD
Abbreviations: DSC, dice similarity coefficient; HD, Hausdorff distance; IoU, intersection over union.aParameter settings in the prior literature [61].
Baseline××0.316 (0.296–0.336)0.441 (0.417–0.465)4.799 (4.681–4.917)
Ours15 × 1510 × 100.331 (0.311–0.351)0.458 (0.434–0.482)4.691 (4.603–4.779)
15 × 150.338 (0.316–0.360)0.462 (0.438–0.486)4.439 (4.343–4.535)
20 × 200.338 (0.318–0.358)0.463 (0.439–0.487)4.538 (4.438–4.638)
19 × 19a10 × 100.332 (0.312–0.352)0.455 (0.431–0.479)4.521 (4.435–4.607)
15 × 15a0.336 (0.316–0.356)0.461 (0.437–0.485)4.558 (4.454–4.662)
20 × 200.340 (0.320–0.360)0.465 (0.441–0.489)4.488 (4.374–4.602)
25 × 2510 × 100.339 (0.317–0.361)0.463 (0.436–0.490)4.442 (4.364–4.520)
15 × 150.337 (0.319–0.355)0.465 (0.441–0.489)4.570 (4.482–4.658)
20 × 200.346 (0.324–0.368)0.473 (0.448–0.498)4.372 (4.280–4.464)

Robustness study on cover rates in the lung + space discriminator.

MethodsCover rateIoUDSCHD
bbreviations: DSC, dice similarity coefficient; HD, Hausdorff distance; IoU, intersection over union.
Baseline×0.316 (0.296–0.336)0.441 (0.417–0.465)4.799 (4.681–4.917)
Ours0.800.334 (0.314–0.354)0.460 (0.436–0.484)4.408 (4.306–4.510)
0.900.343 (0.321–0.365)0.470 (0.445–0.495)4.474 (4.380–4.568)
0.990.336 (0.316–0.356)0.461 (0.437–0.485)4.558 (4.454–4.662)