SLIDE 1
Breast cancer screening with multiple views
screening mammography
Improving the Ability of Deep Neural Networks to Use Information - - PowerPoint PPT Presentation
Improving the Ability of Deep Neural Networks to Use Information - - PowerPoint PPT Presentation
Improving the Ability of Deep Neural Networks to Use Information from Multiple Views in Breast Cancer Screening Nan Wu , Stanis aw Jastrz bski, Jungkyu Park, Linda Moy, Kyunghyun Cho, Krzysztof J. Geras Breast cancer screening with multiple
SLIDE 2
SLIDE 3
Breast cancer screening with multiple views
screening mammography
SLIDE 4
Breast cancer screening with multiple views
screening mammography Medio Lateral Oblique (MLO) Cranio Caudal (CC)
SLIDE 5
Research question
SLIDE 6
Research question
Using both views is essential to make an accurate diagnosis in breast cancer screening. Radiologists
SLIDE 7
Research question
Using both views is essential to make an accurate diagnosis in breast cancer screening. Radiologists Does it utilize information in both views? Multiview deep neural networks
CC MLO Classifier DNN DNN Fusion operation
SLIDE 8
Evidence of difficulties in multiview learning
SLIDE 9
Evidence of difficulties in multiview learning
[1] Nan Wu et al. Deep neural networks improve radiologists’ performance in breast cancer screening. IEEE Transactions
- n Medical Imaging, 2019.
SLIDE 10
Image-wise network Joint network
Evidence of difficulties in multiview learning
[1] Nan Wu et al. Deep neural networks improve radiologists’ performance in breast cancer screening. IEEE Transactions
- n Medical Imaging, 2019.
SLIDE 11
Image-wise network Joint network
Evidence of difficulties in multiview learning
[1] Nan Wu et al. Deep neural networks improve radiologists’ performance in breast cancer screening. IEEE Transactions
- n Medical Imaging, 2019.
SLIDE 12
Image-wise network Joint network
Evidence of difficulties in multiview learning
[1] Nan Wu et al. Deep neural networks improve radiologists’ performance in breast cancer screening. IEEE Transactions
- n Medical Imaging, 2019.
SLIDE 13
Image-wise network Joint network
Evidence of difficulties in multiview learning
AUC joint Image-wise Image-only 0.822 0.830 Image-and-heatmaps 0.860 0.875
[1] Nan Wu et al. Deep neural networks improve radiologists’ performance in breast cancer screening. IEEE Transactions
- n Medical Imaging, 2019.
SLIDE 14
Image-wise network Joint network
Evidence of difficulties in multiview learning
AUC joint Image-wise Image-only 0.822 0.830 Image-and-heatmaps 0.860 0.875
[1] Nan Wu et al. Deep neural networks improve radiologists’ performance in breast cancer screening. IEEE Transactions
- n Medical Imaging, 2019.
[2] Weiyao Wang, Du Tran, and Matt Feiszli. What makes training multi-modal networks hard? arXiv:1905.12681, 2019.
[3] Mohammad Hashir, Hadrien Bertrand, Joseph Paul Cohen. Quantifying the Value of Lateral Views in Deep Learning for
Chest X-rays. MIDL 2020.
SLIDE 15
What makes using both views of the breast difficult?
SLIDE 16
What makes using both views of the breast difficult?
CC MLO Classifier DNN Classifier DNN
SLIDE 17
What makes using both views of the breast difficult?
CC MLO Classifier DNN Classifier DNN
CC MLO Classifier DNN DNN Fusion operation
SLIDE 18
How to improve its ability in utilizing information in both views of the breast?
SLIDE 19
How to improve its ability in utilizing information in both views of the breast?
+ Two methods that can help: Modality Dropout and Sharing weights between part operating on each view.
SLIDE 20
Thank you!
Improving the Ability of Deep Neural Networks to Use Information from Multiple Views in Breast Cancer Screening
SLIDE 21
Improving the Ability of Deep Neural Networks to Use Information from Multiple Views in Breast Cancer Screening
Nan Wu, Stanisław Jastrzębski, Jungkyu Park, Linda Moy, Kyunghyun Cho, Krzysztof J. Geras
CC MLO Classifier DNN DNN Fusion operation
- What makes using both views of the breast
difficult?
- How to improve its ability in utilizing
information in both views of the breast?
Medio Lateral Oblique Cranio Caudal
screening mammography