DeepMiner: Discovering Interpretable Representations for Mammogram Classification and Explanation
By Jimmy Wu, Bolei Zhou, Diondra Peck, Scott S. Hsieh, Vandana Dialani, Lester W. Mackey, Genevieve Patterson. 2018
25 citations recorded by the source corpus.
We propose DeepMiner, a framework to discover interpretable representations in deep neural networks and to build explanations for medical predictions. By probing convolutional neural networks (CNNs) trained to classify cancer in mammograms, we show that many individual units in the final convolutional layer of a CNN respond strongly to diseased tissue concepts specified by the BI-RADS lexicon. After expert annotation of the interpretable units, our proposed method is able to generate explanations for CNN mammogram classification that are correlated with ground truth radiology reports on the DDSM dataset. We show that DeepMiner not only enables better understanding of the nuances of CNN classification decisions, but also possibly discovers new visual knowledge relevant to medical diagnosis.
Subfields: Artificial Intelligence, Radiology, Nuclear Medicine and Imaging
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Mohammad Rostami
- Soheil Kolouri
- Cynthia Rudin
- Alina Jade Barnett
- Chaofan Tao
- Heiko Hoffmann
- Phillip E. Pope
- Charles E. Martin
In Radiology, Nuclear Medicine and Imaging
Citations in the corpus
- A case-based interpretable deep learning model for classification of mass lesions in digital mammography (2021, methodology): “Ours [54] [27] [53] Inherently interpretable model (not posthoc) X X X Provides global interpretability (on model) X X Provides local interpretability (on each case) X X X Explanation is guaranteed to match model reasoning X Incorporate domain-specific terminology X X X Provides similar prototypes f”
- Explainability Methods for Graph Convolutional Neural Networks (2019, background): “For example, in the area of medical imaging [34], in addition to classifying images having malignant lesions, they can be localized, as the CNN can provide reasoning for classifying an input image.”