Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
194 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
46 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Improving deep neural network generalization and robustness to background bias via layer-wise relevance propagation optimization (2202.00232v7)

Published 1 Feb 2022 in eess.IV, cs.CV, and cs.LG

Abstract: Features in images' backgrounds can spuriously correlate with the images' classes, representing background bias. They can influence the classifier's decisions, causing shortcut learning (Clever Hans effect). The phenomenon generates deep neural networks (DNNs) that perform well on standard evaluation datasets but generalize poorly to real-world data. Layer-wise Relevance Propagation (LRP) explains DNNs' decisions. Here, we show that the optimization of LRP heatmaps can minimize the background bias influence on deep classifiers, hindering shortcut learning. By not increasing run-time computational cost, the approach is light and fast. Furthermore, it applies to virtually any classification architecture. After injecting synthetic bias in images' backgrounds, we compared our approach (dubbed ISNet) to eight state-of-the-art DNNs, quantitatively demonstrating its superior robustness to background bias. Mixed datasets are common for COVID-19 and tuberculosis classification with chest X-rays, fostering background bias. By focusing on the lungs, the ISNet reduced shortcut learning. Thus, its generalization performance on external (out-of-distribution) test databases significantly surpassed all implemented benchmark models.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (51)
  1. Geirhos, R. et al. Shortcut learning in deep neural networks. \JournalTitleNature Machine Intelligence 2, 665–673, DOI: 10.1038/s42256-020-00257-z (2020).
  2. Bach, S. et al. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. \JournalTitlePLOS ONE 10, 1–46, DOI: 10.1371/journal.pone.0130140 (2015).
  3. Ouyang, X. et al. Learning hierarchical attention for weakly-supervised chest x-ray abnormality localization and diagnosis. \JournalTitleIEEE Transactions on Medical Imaging 40, 2698–2710, DOI: 10.1109/TMI.2020.3042773 (2021).
  4. Tell me where to look: Guided attention inference network. DOI: 10.1109/CVPR.2018.00960 (2018).
  5. Selvaraju, R. R. et al. Grad-cam: Visual explanations from deep networks via gradient-based localization. In 2017 IEEE International Conference on Computer Vision (ICCV), 618–626, DOI: 10.1109/ICCV.2017.74 (2017).
  6. Right for the right reasons: Training differentiable models by constraining their explanations. In Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI-17, 2662–2670, DOI: 10.24963/ijcai.2017/371 (2017).
  7. Deep inside convolutional networks: Visualising image classification models and saliency maps. \JournalTitlearXiv 1312.6034 (2014).
  8. Not just a black box: Learning important features through propagating activation differences. \JournalTitleArXiv abs/1605.01713 (2016).
  9. Schlemper, J. et al. Attention gated networks: Learning to leverage salient regions in medical images, DOI: 10.48550/ARXIV.1808.08114 (2018).
  10. Dosovitskiy, A. et al. An image is worth 16x16 words: Transformers for image recognition at scale (2020). 2010.11929.
  11. Covid-19 image data collection. \JournalTitlearXiv 2003.11597 (2020).
  12. Signoroni, A. et al. Bs-net: learning covid-19 pneumonia severity on a large chest x-ray dataset. \JournalTitleMedical Image Analysis 71, 102046, DOI: 10.1016/j.media.2021.102046 (2021).
  13. A critic evaluation of methods for covid-19 automatic detection from x-ray images (2020). arXiv:2004.12823.
  14. Current limitations to identify covid-19 using artificial intelligence with chest x-ray imaging (part ii). the shortcut learning problem. \JournalTitleHealth and Technology 11, DOI: 10.1007/s12553-021-00609-8 (2021).
  15. Ai for radiographic covid-19 detection selects shortcuts over signal. \JournalTitleNat Mach Intell 3, 610–619, DOI: 10.1038/s42256-021-00338-7 (2021).
  16. Covid-19 detection using chest x-rays: is lung segmentation important for generalization? (2021). 2104.06176.
  17. Teixeira, L. O. et al. Impact of lung segmentation on the diagnosis and explanation of covid-19 in chest x-ray images. \JournalTitleSensors 21, DOI: 10.3390/s21217116 (2021).
  18. Guan, W.-j. et al. Clinical characteristics of coronavirus disease 2019 in china. \JournalTitleNew England Journal of Medicine 382, 1708–1720, DOI: 10.1056/NEJMoa2002032 (2020). https://doi.org/10.1056/NEJMoa2002032.
  19. Kim, E. A. et al. Viral pneumonias in adults: Radiologic and pathologic findings. \JournalTitleRadioGraphics 22, S137–S149, DOI: 10.1148/radiographics.22.suppl_1.g02oc15s137 (2002). PMID: 12376607.
  20. Rosenthal, A. et al. The tb portals: an open-access, web-based platform for global drug-resistant-tuberculosis data sharing and analysis. \JournalTitleJournal of Clinical Microbiology 55, JCM.01013–17, DOI: 10.1128/JCM.01013-17 (2017).
  21. A systematic review of deep learning techniques for tuberculosis detection from chest radiograph. \JournalTitleFrontiers in Medicine 9, DOI: 10.3389/fmed.2022.830515 (2022).
  22. Deep learning for automated classification of tuberculosis-related chest x-ray: dataset distribution shift limits diagnostic performance generalizability. \JournalTitleHeliyon 6, e04614, DOI: https://doi.org/10.1016/j.heliyon.2020.e04614 (2020).
  23. Rahman, T. et al. Reliable tuberculosis detection using chest x-ray with deep learning, segmentation and visualization. \JournalTitleIEEE Access 8, 191586–191601, DOI: 10.1109/ACCESS.2020.3031384 (2020).
  24. Organization, W. H. Chest radiography in tuberculosis detection: summary of current WHO recommendations and guidance on programmatic approaches (World Health Organization, 2016).
  25. Irvin, J. et al. Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison. \JournalTitleProceedings of the AAAI Conference on Artificial Intelligence 33, 590–597, DOI: 10.1609/aaai.v33i01.3301590 (2019).
  26. Jaeger, S. et al. Two public chest x-ray datasets for computer-aided screening of pulmonary diseases. \JournalTitleQuantitative imaging in medicine and surgery 4 6, 475–7, DOI: 10.3978/j.issn.2223-4292.2014.11.20 (2014).
  27. Deep learning face attributes in the wild. In 2015 IEEE International Conference on Computer Vision (ICCV), 3730–3738, DOI: 10.1109/ICCV.2015.425 (2015).
  28. Novel dataset for fine-grained image categorization. In First Workshop on Fine-Grained Visual Categorization, IEEE Conference on Computer Vision and Pattern Recognition (Colorado Springs, CO, 2011).
  29. The surprising impact of mask-head architecture on novel class segmentation (2021). 2104.00613.
  30. Improving deep neural network generalization and robustness to background bias via layer-wise relevance propagation optimization. \JournalTitleGitHub DOI: https://doi.org/10.5281/zenodo.8408250 (2023).
  31. Very deep convolutional networks for large-scale image recognition (2014). 1409.1556.
  32. U-net: Convolutional networks for biomedical image segmentation. In Navab, N., Hornegger, J., Wells, W. M. & Frangi, A. F. (eds.) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, 234–241 (Springer International Publishing, Cham, 2015).
  33. Densely connected convolutional networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2261–2269, DOI: 10.1109/CVPR.2017.243 (2017).
  34. Bressem, K. K. et al. Comparing different deep learning architectures for classification of chest radiographs. \JournalTitleScientific Reports 10, DOI: 10.1038/s41598-020-70479-z (2020).
  35. A simple generalisation of the area under the roc curve for multiple class classification problems. \JournalTitleMach. Learn. 45, 171–186, DOI: 10.1023/A:1010920819831 (2001).
  36. A deep convolutional neural network for covid-19 detection using chest x-rays. \JournalTitleResearch on Biomedical Engineering DOI: 10.1007/s42600-021-00132-9 (2021).
  37. Supplementary data for "improving deep neural network generalization and robustness to background bias via layer-wise relevance propagation optimization". \JournalTitleFigshare DOI: https://doi.org/10.6084/m9.figshare.24243895.v1 (2023).
  38. Explain and improve: Lrp-inference fine-tuning for image captioning models. \JournalTitleInformation Fusion 77, 233–246, DOI: https://doi.org/10.1016/j.inffus.2021.07.008 (2022).
  39. Testing the robustness of attribution methods for convolutional neural networks in mri-based alzheimer’s disease classification. In Suzuki, K. et al. (eds.) Interpretability of Machine Intelligence in Medical Image Computing and Multimodal Learning for Clinical Decision Support, 3–11 (Springer International Publishing, Cham, 2019).
  40. Towards better understanding of gradient-based attribution methods for deep neural networks (2018). 1711.06104.
  41. Layer-Wise Relevance Propagation: An Overview, 193–209 (Springer International Publishing, Cham, 2019).
  42. Explaining nonlinear classification decisions with deep taylor decomposition. \JournalTitlePattern Recognition 65, 211–222, DOI: 10.1016/j.patcog.2016.11.008 (2017).
  43. Qiu, S. Global weighted average pooling bridges pixel-level localization and image-level classification, DOI: 10.48550/ARXIV.1809.08264 (2018).
  44. Wang, X. et al. Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. \JournalTitle2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) DOI: 10.1109/cvpr.2017.369 (2017).
  45. de la Iglesia Vayá, M. et al. Bimcv covid-19+: a large annotated dataset of rx and ct images from covid-19 patients (2020). 2006.01174.
  46. Johnson, A. et al. Mimic-cxr-jpg - chest radiographs with structured labels (version 2.0.0). https://doi.org/10.13026/8360-t248 (2019).
  47. Johnson, A. E. W. et al. Mimic-cxr-jpg, a large publicly available database of labeled chest radiographs, DOI: 10.48550/ARXIV.1901.07042 (2019).
  48. Goldberger, A. et al. Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals. \JournalTitleCirculation 101, E215—20, DOI: 10.1161/01.cir.101.23.e215 (2000).
  49. Guo, M.-H. et al. Attention mechanisms in computer vision: A survey. \JournalTitleComputational Visual Media DOI: 10.1007/s41095-022-0271-y (2022).
  50. Recurrent models of visual attention. In Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2, NIPS’14, 2204–2212 (MIT Press, Cambridge, MA, USA, 2014).
  51. Spatial transformer networks, DOI: 10.48550/ARXIV.1506.02025 (2015).
Citations (19)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now