Source code for Microsoft Code Fun Do Hackathon 2018 organized at IIT (BHU), Varanasi.
To assist radiologists using deep learning. We aim to build a abnormality detection model to detect and predict abnormality in musculoskeletal radiographs, such a model could be utilized for worklist prioritization. In this scenario, the studies detected as abnormal could be moved ahead in the image interpretation workflow, allowing the sickest patients to receive quicker diagnoses and treatment.
We developed a 169 layer Dense Convolutional Neural Network (DenseNet) for predicting abnormality on musculoskeletal radiographs. We developed a Django webserver with a basic user interface to upload and test radiographic images on our model. The Django site was hosted on a Microsof Azure Ubuntu virtual machine.
The model is a 169 layer DenseNet with single node output layer initialized with weights from a model pretrained on ImageNet dataset. Before feeding the images to the network, each image is scaled to 224 x 224, normalized to have same mean and standard deviation as of the images in the ImageNet training set. The model takes mean of predictions on all the images of a study to predict the overall abnormality of the corresponding study. The model was trained on MURA dataset, achieving 79.41% accuracy on validation set.
To know more about training process: DenseNet-MURA-PyTorch.
Trained model state is stored here. The model was trained on WRIST study type only, due to time constraints.
- PyTorch 0.3
- Django 1.11
- Pillow 5.0.0
- Install the dependencies by
pip install -r requirements.txt - run the server with
python manage.py runserver
- Hemanth Sai
- Srimukha Paturi
- Myself :)
Won third prize among ~75 teams.:)
Feel free to raise an issue to know more about this project.