EyeAI - Android app

1 minute read

Github link to project

Prototype Deep leanring based android app for eye disease on edge devices

Logo

Aim

Training, deploying deep learning models for ophthalmology diagnosis on android app

Editing a markdown file for a talk

Get it on Google Play

Goals -

  • 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%

Screenshots

| Normal | Diabetic retinopathy | Maculopathy | Retinitis pigmentosa | | :—: | :—: | :—: | :—: | | Normal | Maculopathy | Diabetic retinopathy | Retinitis pigmentosa |

False image type prediction

| CT scans | Xrays | Random faces | Screenshots | | :—: | :—: | :—: | :—: | | CT_scan | Xrays | Random faces | Screenshots |

Training

| Training loss | Training accuracy | Confusion matrix | | :—: | :—: | :—: |
| Training loss | Training accuracy | Confusion matrix |

Maybe sometime in future

  • 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