Liver and Tumor Segmentation for CT Images with U-Net
- This is the final project for MIT 6.869 Spring 2022.
- The source code is available on my github repo.
Method
- Network structure: U-Net with 16 features on the first layer. Optimizer was Adam. Loss function was crossEntropy
- Data set: liver data set from Medical Segmentation Decathlon. I sliced the 3-D CT images into 3-channel 2-D images for training and validation.
- An experimental idea of fusing the prediction labels along the three directions.
Result
 |
 |
 |
 |
|---|---|---|---|
| CT image | True label | Prediction | Fused prediction |