User Stories
As a potential client I want to load in unlabeled brain MRI scans into the model.
As a potential client I want to be able to get segmented brain MRI imaging from the model.
As a potential client I want the output images to be segmented by blood, skin, bone, brain matter, air, and cerebral spine fluid.
As a potential client I want to be able to take the output images and input them into another classifier model as training data.
As a potential client I want these segmented images to be highly accurate so the training data is efficient and correct.
Specifications
Functional
- Definite
- Trained GAN model that successfully segments Brain MRI scans into 5 categories: blood, skin, bone, brain matter, air, and cerebral spinal fluid.
- Tested GAN model for accuracy and made adjustments to improve model
- Model backend will be properly integrated with frontend web app
- Segmented MRI images should be able to be taken into classifier model as training data
- Image pre-processing and normalization to occur prior to inputting data into the model.
- Formally evaluated model with ROC curves to provide metrics of performance
- Perhaps
- Model will be able to function for both CT and MRI imaging
- Implement both binary and multi-class models
- Allow user to download the image results from model code
- Improbable
- Grid search implementation for parameter tuning
- Functional neurosurgery frailty score – automated prediction of outcomes based on anatomic features
Non-Functional
- Definite
- Users can upload input images to run in model
- Results will be concisely displayed on web app
- Frontend should be both mobile and desktop compatible
- Perhaps
- If wanted by client, implement user authentication for usage
- Mobile app integration on iOS
- User will be able to download image results from frontend