Advanced search
1 file | 2.56 MB Add to list

Towards a quantitative analysis of class activation mapping for deep learning-based computer-aided diagnosis

Hanul Kang (UGent) , Homin Park (UGent) , Yuju Ahn, Arnout Van Messem (UGent) and Wesley De Neve (UGent)
Author
Organization
Abstract
Class Activation Mapping (CAM) can be used to obtain a visual understanding of the predictions made by Convolutional Neural Networks (CNNs), facilitating qualitative insight into these neural networks when they are, for instance, used for the purpose of medical image analysis. In this paper, we investigate to what extent CAM also enables a quantitative understanding of CNN-based classification models through the creation of segmentation masks out of class activation maps, hereby targeting the use case of brain tumor classification. To that end, when a class activation map has been created for a correctly classified brain tumor, we additionally perform tumor segmentation by binarization of the aforementioned map, leveraging different methods for thresholding. In a next step, we compare this CAM-based segmentation mask to the segmentation ground truth, measuring similarity through the use of Intersection over Union (IoU). Our experimental results show that, although our CNN-based classification models have a similarly high accuracy between 86.0% and 90.8%, their generated masks are different. For example, our Modified VGG-16 model scores an mIoU of 12.2%, whereas AlexNet scores an mIoU of 2.1%. When comparing with the mIoU obtained by our U-Net-based models, which is between 66.6% and 67.3%, and where U-Net is a dedicated pixel-wise segmentation model, our experimental results point to a significant difference in terms of segmentation effectiveness. As such, the use of CAM for the purpose of proxy segmentation or as a ground truth segmentation mask generator comes with several limitations.

Downloads

  • DS401.pdf
    • full text (Published version)
    • |
    • open access
    • |
    • PDF
    • |
    • 2.56 MB

Citation

Please use this url to cite or link to this publication:

MLA
Kang, Hanul, et al. “Towards a Quantitative Analysis of Class Activation Mapping for Deep Learning-Based Computer-Aided Diagnosis.” MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT, edited by Frank W. Samuelson and Sian Taylor-Phillips, vol. 11599, SPIE, 2021, doi:10.1117/12.2580819.
APA
Kang, H., Park, H., Ahn, Y., Van Messem, A., & De Neve, W. (2021). Towards a quantitative analysis of class activation mapping for deep learning-based computer-aided diagnosis. In F. W. Samuelson & S. Taylor-Phillips (Eds.), MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT (Vol. 11599). Online: SPIE. https://doi.org/10.1117/12.2580819
Chicago author-date
Kang, Hanul, Homin Park, Yuju Ahn, Arnout Van Messem, and Wesley De Neve. 2021. “Towards a Quantitative Analysis of Class Activation Mapping for Deep Learning-Based Computer-Aided Diagnosis.” In MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT, edited by Frank W. Samuelson and Sian Taylor-Phillips. Vol. 11599. SPIE. https://doi.org/10.1117/12.2580819.
Chicago author-date (all authors)
Kang, Hanul, Homin Park, Yuju Ahn, Arnout Van Messem, and Wesley De Neve. 2021. “Towards a Quantitative Analysis of Class Activation Mapping for Deep Learning-Based Computer-Aided Diagnosis.” In MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT, ed by. Frank W. Samuelson and Sian Taylor-Phillips. Vol. 11599. SPIE. doi:10.1117/12.2580819.
Vancouver
1.
Kang H, Park H, Ahn Y, Van Messem A, De Neve W. Towards a quantitative analysis of class activation mapping for deep learning-based computer-aided diagnosis. In: Samuelson FW, Taylor-Phillips S, editors. MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT. SPIE; 2021.
IEEE
[1]
H. Kang, H. Park, Y. Ahn, A. Van Messem, and W. De Neve, “Towards a quantitative analysis of class activation mapping for deep learning-based computer-aided diagnosis,” in MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT, Online, 2021, vol. 11599.
@inproceedings{8694842,
  abstract     = {{Class Activation Mapping (CAM) can be used to obtain a visual understanding of the predictions made by Convolutional Neural Networks (CNNs), facilitating qualitative insight into these neural networks when they are, for instance, used for the purpose of medical image analysis. In this paper, we investigate to what extent CAM also enables a quantitative understanding of CNN-based classification models through the creation of segmentation masks out of class activation maps, hereby targeting the use case of brain tumor classification. To that end, when a class activation map has been created for a correctly classified brain tumor, we additionally perform tumor segmentation by binarization of the aforementioned map, leveraging different methods for thresholding. In a next step, we compare this CAM-based segmentation mask to the segmentation ground truth, measuring similarity through the use of Intersection over Union (IoU). Our experimental results show that, although our CNN-based classification models have a similarly high accuracy between 86.0% and 90.8%, their generated masks are different. For example, our Modified VGG-16 model scores an mIoU of 12.2%, whereas AlexNet scores an mIoU of 2.1%. When comparing with the mIoU obtained by our U-Net-based models, which is between 66.6% and 67.3%, and where U-Net is a dedicated pixel-wise segmentation model, our experimental results point to a significant difference in terms of segmentation effectiveness. As such, the use of CAM for the purpose of proxy segmentation or as a ground truth segmentation mask generator comes with several limitations.}},
  articleno    = {{115990M}},
  author       = {{Kang, Hanul and Park, Homin and Ahn, Yuju and Van Messem, Arnout and De Neve, Wesley}},
  booktitle    = {{MEDICAL IMAGING 2021 : IMAGE PERCEPTION, OBSERVER PERFORMANCE, AND TECHNOLOGY ASSESSMENT}},
  editor       = {{Samuelson, Frank W. and Taylor-Phillips, Sian}},
  isbn         = {{9781510640276}},
  issn         = {{0277-786X}},
  language     = {{eng}},
  location     = {{Online}},
  pages        = {{13}},
  publisher    = {{SPIE}},
  title        = {{Towards a quantitative analysis of class activation mapping for deep learning-based computer-aided diagnosis}},
  url          = {{http://dx.doi.org/10.1117/12.2580819}},
  volume       = {{11599}},
  year         = {{2021}},
}

Altmetric
View in Altmetric
Web of Science
Times cited: