Abstract
Image scale carries crucial information in medical imaging, e.g. the size and spatial frequency of local structures, lesions, tumors and cell nuclei. With feature transfer being a common practice, scale-invariant features implicitly learned from pretraining on ImageNet tend to be preferred over scale-covariant features. The pruning strategy in this paper proposes a way to maintain scale covariance in the transferred features. Deep learning interpretability is used to analyze the layer-wise encoding of scale information for popular architectures such as InceptionV3 and ResNet50. Interestingly, the covariance of scale peaks at central layers and decreases close to softmax. Motivated by these results, our pruning strategy removes the layers where invariance to scale is learned. The pruning operation leads to marked improvements in the regression of both nuclei areas and magnification levels of histopathology images. These are relevant applications to enlarge the existing medical datasets with open-access images as those of PubMed Central. All experiments are performed on publicly available data and the code is shared on GitHub.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Downloadable at https://keras.io/api/applications/.
- 2.
- 3.
Following the same terminology, the equivariance, as opposed to covariance, implies that the function \(\phi (\cdot )\) maps an input image to a point in the same domain, i.e. \(\phi : \mathbb {R}^{h\times w} \rightarrow \mathbb {R}^{h \times w}\).
- 4.
For simplicity, we omit the intercept. In Eq. (1), the intercept would be \(v_0\) with \(\phi _0(g_{\sigma }(\mathrm {X}))=1\).
- 5.
We compute \(R^2=\frac{\sum _{i=1}^N (\hat{r}_i-\bar{r})}{\sum _{i=1}^N r_i - \bar{r}}\), were N is the number of test data samples, \(\hat{r}\) is the ratio predicted by the regression model, \(\bar{r}\) is the mean of the true ratios \(\{r_i\}_{i=1}^N\).
- 6.
- 7.
Layer names refer to the Keras implementation names.
- 8.
Different seeds were used to initialize the dense connections to the last dense layer.
References
Andrearczyk, V., Fageot, J., Oreiller, V., Montet, X., Depeursinge, A.: Exploring local rotation invariance in 3D CNNs with steerable filters. In: International Conference on Medical Imaging with Deep Learning (2019)
Andrearczyk, V., Graziani, M., Müller, H., Depeursinge, A.: Consistency of scale equivariance in internal representations of CNNs. In: Irish Machine Vision and Image Processing (2020)
Bejnordi, B.E., et al.: Context-aware stacked convolutional neural networks for classification of breast carcinomas in whole-slide histopathology images. J. Med. Imaging 4(4), 044504 (2017)
Bruna, J., Mallat, S.: Invariant scattering convolution networks. IEEE Trans. Pattern Anal. Mach. Intell. 35(8), 1872–1886 (2013)
Cai, C.J., et al.: Human-centered tools for coping with imperfect algorithms during medical decision-making. In: Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, pp. 1–14 (2019)
Depeursinge, A., Foncubierta-Rodriguez, A., Van De Ville, D., Müller, H.: Three-dimensional solid texture analysis in biomedical imaging: review and opportunities. Med. Image Anal. 18(1), 176–196 (2014)
Depeursinge, A.: Multi-scale and multi-directional biomedical texture analysis: finding the needle in the haystack. In: Biomedical Texture Analysis: Fundamentals. Applications and Tools, Elsevier-MICCAI Society Book Series, pp. 29–53. Elsevier (2017)
Elston, C.W., Ellis, I.O.: Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: experience from a large study with long-term follow-up. Histopathology 19(5), 403–410 (1991)
Everingham, M., Van Gool, L., Williams, C.K., Winn, J., Zisserman, A.: The pascal visual object classes (VOC) challenge. Int. J. Comput. Vision 88(2), 303–338 (2010)
Ghosh, R., Gupta, A.K.: Scale steerable filters for locally scale-invariant convolutional neural networks. In: Workshop on Theoretical Physics for Deep Learning at the International Conference on Machine Learning (2019)
Graziani, M., Andrearczyk, V., Marchand-Maillet, S., Müller, H.: Concept attribution: explaining CNN decisions to physicians. Comput. Biol. Med. 123, 103865 (2020)
Graziani, M., Andrearczyk, V., Müller, H.: Regression concept vectors for bidirectional explanations in histopathology. In: Stoyanov, D., et al. (eds.) MLCN/DLF/IMIMIC-2018. LNCS, vol. 11038, pp. 124–132. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-02628-8_14
Graziani, M., Andrearczyk, V., Müller, H.: Visualizing and interpreting feature reuse of pretrained CNNs for histopathology. In: Irish Machine Vision and Image Processing (IMVIP) (2019)
Graziani, M., Müller, H., Andrearczyk, V.: Interpreting intentionally flawed models with linear probes. In: SDL-CV Workshop at the IEEE International Conference on Computer Vision (2019)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778 (2016)
Hu, Z., Tang, J., Wang, Z., Zhang, K., Zhang, L., Sun, Q.: Deep learning for image-based cancer detection and diagnosis-a survey. Pattern Recogn. 83, 134–149 (2018)
Huh, M., Agrawal, P., Efros, A.A.: What makes ImageNet good for transfer learning? In: Workshop on Large Scale Computer Vision Systems at NeurIPS 2016 (2016)
Janowczyk, A., Madabhushi, A.: Deep learning for digital pathology image analysis: a comprehensive tutorial with selected use cases. J. Pathol. Inform. 7 (2016)
Kanazawa, A., Sharma, A., Jacobs, D.W.: Locally scale-invariant convolutional neural networks. In: Advances in Neural Information Processing Systems (2014)
Kim, B., et al.: Interpretability beyond feature attribution: quantitative testing with concept activation vectors (TCAV). In: International Conference on Machine Learning, pp. 2673–2682 (2018)
Litjens, G., et al.: A survey on deep learning in medical image analysis. Med. Image Anal. 42, 60–88 (2017)
Lompech, T., Graziani, M., Otálora, S., Depeursinge, A., Andrearczyk, V.: On the scale invariance in state of the art CNNs trained on ImageNet (2020, submitted)
Marcos, D., Kellenberger, B., Lobry, S., Tuia, D.: Scale equivariance in CNNs with vector fields. In: FAIM workshop at the International Conference on Machine Learning (2018)
Otálora, S., Atzori, M., Andrearczyk, V., Müller, H.: Image magnification regression using DenseNet for exploiting histopathology open access content. In: Stoyanov, D., et al. (eds.) OMIA/COMPAY-2018. LNCS, vol. 11039, pp. 148–155. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-00949-6_18
Raghu, M., Zhang, C., Kleinberg, J., Bengio, S.: Transfusion: understanding transfer learning with applications to medical imaging. arXiv preprint arXiv:1902.07208 (2019)
Shin, H.C., et al.: Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning. IEEE Trans. Med. Imaging 35(5), 1285–1298 (2016)
Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., Wojna, Z.: Rethinking the inception architecture for computer vision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2818–2826 (2016)
Szeliski, R.: Computer Vision: Algorithms and Applications. Springer, London (2010). https://doi.org/10.1007/978-1-84882-935-0
Touvron, H., Vedaldi, A., Douze, M., Jégou, H.: Fixing the train-test resolution discrepancy. In: Advances in Neural Information Processing Systems (2019)
Van Noord, N., Postma, E.: Learning scale-variant and scale-invariant features for deep image classification. Pattern Recogn. 61, 583–592 (2017)
Veeling, B.S., Linmans, J., Winkens, J., Cohen, T., Welling, M.: Rotation equivariant CNNs for digital pathology. In: Frangi, A.F., Schnabel, J.A., Davatzikos, C., Alberola-López, C., Fichtinger, G. (eds.) MICCAI 2018. LNCS, vol. 11071, pp. 210–218. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-00934-2_24
Wan, T., Cao, J., Chen, J., Qin, Z.: Automated grading of breast cancer histopathology using cascaded ensemble with combination of multi-level image features. Neurocomputing 229, 34–44 (2017)
Worrall, D.E., Welling, M.: Deep scale-spaces: equivariance over scale. arXiv preprint arXiv:1905.11697 (2019)
Yan, E., Huan, Y.: Do CNNs encode data augmentations? arxiv.org/2003.08773 (2020)
Yosinski, J., Clune, J., Bengio, Y., Lipson, H.: How transferable are features in deep neural networks? In: Advances in Neural Information Processing Systems, pp. 3320–3328 (2014)
Acknowledgements
This work was partially possible thanks to the project PROCESS, part of the European Union’s Horizon 2020 research and innovation program (grant agreement No 777533). This work was also supported by the Swiss National Science Foundation (grant 205320_179069).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Graziani, M., Lompech, T., Müller, H., Depeursinge, A., Andrearczyk, V. (2020). Interpretable CNN Pruning for Preserving Scale-Covariant Features in Medical Imaging. In: Cardoso, J., et al. Interpretable and Annotation-Efficient Learning for Medical Image Computing. IMIMIC MIL3ID LABELS 2020 2020 2020. Lecture Notes in Computer Science(), vol 12446. Springer, Cham. https://doi.org/10.1007/978-3-030-61166-8_3
Download citation
DOI: https://doi.org/10.1007/978-3-030-61166-8_3
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-61165-1
Online ISBN: 978-3-030-61166-8
eBook Packages: Computer ScienceComputer Science (R0)