This Tensorflow Model can classify Strabismus eyes and No Strabismus Eyes as normal by Image of eyes.
The model has 3 convolutional layers with max pooling between them. layers use 64, 96 and 128 filters respectively. A flatten layer comes after these convolutional layers to flat the created 3d tensors. Then the extracted features gave to a 256 units fully connected layer followed by another fully connected layers that has 128 units. After that a sigmoid fully connected layer placed.
Results on the test data after training: (Normal Mode)
| Measure | Value |
|---|---|
| Balanced Accuracy | 82% |
| Sensitivity | 86% |
| Specificity | 77% |
Results on the test data after training: (Strict Mode) the threshold set to 0.285
| Measure | Value |
|---|---|
| Balanced Accuracy | 79% |
| Sensitivity | 66% |
| Specificity | 92% |
The model gets 23:9 ratio image of eyes as input to detect strabismus. A sample image of input data is given below.
The model can be used in these two ways.
- Use presented function with python.
- Use model in your way. (The trained model has been provided in “model” directory)
To use "strabismus_predict" function give it image path as string.