Models:
Samuel Callahan
SCallahan/
SpineAI-Detect-Classify-L
0
Downloads: 1
Diff of /Classification/README.md [000000] .. [db7631]

Switch to unified view

a b/Classification/README.md 

  1
  
# ROI Classification Model





  
  

  2
  





  
  

  3
  
[![TensorFlow 1.15](https://img.shields.io/badge/TensorFlow-1.15-FF6F00?logo=tensorflow)](https://github.com/tensorflow/tensorflow/releases/tag/v1.15.0)





  
  

  4
  
[![Python 3.6](https://img.shields.io/badge/Python-3.6-3776AB)](https://www.python.org/downloads/release/python-360/)





  
  

  5
  





  
  

  6
  
![train-classifier](../imgs/Rad_Fig_v4_Classifier.png)





  
  

  7
  





  
  

  8
  
## Training Setup





  
  

  9
  





  
  

  10
  
Models are implemented with Tensorflow 1.15 and trained on NVIDIA GeForce RTX/GTX GPU devices with CUDA version 9 or 10.





  
  

  11
  





  
  

  12
  
Classification model is developed with Keras sequential class, consisting of a stack of six convolution layers and two fully-connected layers that end with softmax activation for multi-class classification.





  
  

  13
  





  
  

  14
  
## Weighted Categorical Cross-entropy Loss





  
  

  15
  





  
  

  16
  
We use a cost-sensitive loss for training due to the highly imbalanced data distirbution.





  
  

  17
  





  
  

  18
  
See Keras Github for more detailed discussion on the implementation:





  
  

  19
  
https://github.com/keras-team/keras/issues/2115





  
  

  20
  





  
  

  21
  





  
  

  22
  
## References





  
  

  23
  
- Keras official tutorial on image classification models: https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html





  
  

  24