AIM - Training, deploying deep learning models for ophthalmology diagnosis on android app
- Train models
- Test models
- Android app framework to deploy model
- On device model prediction
- Capture image / Upload image from device
- Image testing before prediction(Is it a Retina image or not?)
- Report False predictions by model
- Add "Share AI's Report" button
- Without images
- With images
- Suggestions to improve app button
- Prediction probabilities
- Choosing model function - Multiple models for different diseases
- Annonymization of files before sending to model
- Image augmentation before making prediction
Frameworks used -
- Model training - Tensorflow lite
- Android deployment - Kotlin and Android Studio
App size - 20 mb Coverage of Android devices - more than 98%
| Normal | Diabetic retinopathy | Maculopathy | Retinitis pigmentosa |
|---|---|---|---|
 |
 |
 |
 |
| CT scans | Xrays | Random faces | Screenshots |
|---|---|---|---|
 |
 |
 |
|
| Training loss | Training accuracy | Confusion matrix |
|---|---|---|
 |
 |
 |
- Uploading folder of images at once
- Cloud based interpretation on bigger/better models
- Links to model training process and accuracy statistics
- GradCam/Explainable AI components