Great news! Our paper "Deep Neural Networks Improve Radiologists' Performance in Breast Cancer Screening" has been published at @IEEE_TMI!
This marks the end of a 3-year long endeavor, at @NYUDataScience and @cai2r, to demonstrate the potential of CNNs on this task. (1/8)
This marks the end of a 3-year long endeavor, at @NYUDataScience and @cai2r, to demonstrate the potential of CNNs on this task. (1/8)
We trained and evaluated neural networks on over 1,000,000 images. Our network achieved an AUC of 0.895 in predicting the presence of cancer in the breast. To validate our model, we conducted a reader study with 14 readers, each reading 720 exams. (2/8)
Our model is as accurate as experienced radiologists when presented with the same data. A hybrid model, averaging the probability of malignancy predicted by a radiologist with a prediction of our neural network, is more accurate than either of the two separately. (3/8)
We aimed for transparency and reproducibility to let others build upon our work.
The paper is available at ieeexplore.ieee.org/document/88613… in open access mode.
The code and the trained models (@PyTorch and @TensorFlow) can be found at github.com/nyukat/breast_…. (4/8)
The paper is available at ieeexplore.ieee.org/document/88613… in open access mode.
The code and the trained models (@PyTorch and @TensorFlow) can be found at github.com/nyukat/breast_…. (4/8)
A report explaining how the data set was extracted in detail is at cs.nyu.edu/~kgeras/report….
We also wrote a blog post explaining our work in a more accessible form:
medium.com/@jasonphang/de… (5/8)
We also wrote a blog post explaining our work in a more accessible form:
medium.com/@jasonphang/de… (5/8)
It was completed by a big team including @NanWu__ @jpatrickpark @zhansheng @ArtieShen @kchonyc @DrLindaMoy @EricKimMD @UjasParikhMD @iwontbecreative @kudkudakpl @zhehuang1 @DrJiyonLee. It has 32 authors! I don't expect to write a paper with as many authors anytime soon! (6/8)
We are now focusing more on methodological advances:
- neural networks better adjusted to high-resolution medical images,
- neural networks producing more interpretable predictions,
- understanding the robustness of neural networks when applied to this task,
- ... (7/8)
- neural networks better adjusted to high-resolution medical images,
- neural networks producing more interpretable predictions,
- understanding the robustness of neural networks when applied to this task,
- ... (7/8)
Do you have a PhD (or soon will) and want to join us in these efforts? Apply for a postdoc! cs.nyu.edu/~kgeras/postdo…
If you did not yet start a PhD, apply to the NYU School of Medicine (med.nyu.edu/research/sackl…) or the NYU Center for Data Science (cds.nyu.edu/academics/phd/). (8/8)
If you did not yet start a PhD, apply to the NYU School of Medicine (med.nyu.edu/research/sackl…) or the NYU Center for Data Science (cds.nyu.edu/academics/phd/). (8/8)
