Abstract:
A deep convolutional neural network for breast cancer screening exam classification was trained and evaluated on over 200000 exams (1000000 images). Our screening population network predicts breast cancer with an AUC of 0.895. Technical advances explain the high accuracy.
- 1) Our network’s novel two-stage architecture and training procedure, which lets us use a high-capacity patch-level network to learn from pixel-level labels and a macroscopic breast-level network.
- 2) A custom ResNet-based network optimized for high-resolution medical images.
- 3) Pretraining the network on noisy label screening BI-RADS classification.
- 4) Selecting the best combination of input views.
Our model is as accurate as experienced radiologists when presented with the same data, according to a reader study with 14 readers reading 720 screening mammogram exams.
We demonstrate that a hybrid model combining a radiologist’s malignancy prediction with our neural network prediction is more accurate than either alone.
We analyze our network’s performance on different screening population subpopulations, the model’s design, training procedure, errors, and internal representations to better understand our results.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
