enhance readme with details of model training

README.md

@@ -11,13 +11,31 @@ A pre-trained model for volumetric (3D) segmentation of the spleen from CT image
 # Model Overview
 This model is trained using the runner-up [1] awarded pipeline of the "Medical Segmentation Decathlon Challenge 2018" using the UNet architecture [2] with 32 training images and 9 validation images.
 
+![image](https://developer.download.nvidia.com/assets/Clara/Images/clara_pt_spleen_ct_segmentation_workflow.png)
+
 ## Data
 The training dataset is Task09_Spleen.tar from http://medicaldecathlon.com/.
 
 ## Training configuration
-The training was performed with at least 12GB-memory GPUs.
+The segmentation of spleen region is formulated as the voxel-wise binary classification. Each voxel is predicted as either foreground (spleen) or background. And the model is optimized with gradient descent method minimizing Dice + cross entropy loss between the predicted mask and ground truth segmentation.
+
+The training was performed with the following:
+
+- GPU: at least 12GB of GPU memory
+- Actual Model Input: 96 x 96 x 96
+- AMP: True
+- Optimizer: Adam
+- Learning Rate: 1e-4
+- Loss: DiceCELoss
 
-Actual Model Input: 96 x 96 x 96
+Pre-processing transforms:
+
+1. Convert data to channel-first
+2. Resample to resolution 1.5 x 1.5 x 2 mm
+3. Scale intensity
+4. Cropping foreground surrounding regions
+5. Cropping random fixed sized regions of size [96,96,96] with the center being a foreground or background voxel at ratio 1 : 1
+6. Randomly shifting intensity of the volume
 
 ## Input and output formats
 Input: 1 channel CT image
@@ -29,6 +47,17 @@ This model achieves the following Dice score on the validation data (our own spl
 
 Mean Dice = 0.96
 
+## Training Performance
+A graph showing the training loss over 1260 epochs (10080 iterations).
+
+![](https://developer.download.nvidia.com/assets/Clara/Images/clara_pt_spleen_ct_segmentation_train_2.png) <br>
+
+## Validation Performance
+A graph showing the validation mean Dice over 1260 epochs.
+
+![](https://developer.download.nvidia.com/assets/Clara/Images/clara_pt_spleen_ct_segmentation_val_2.png) <br>
+
+
 ## commands example
 Execute training:
 

configs/metadata.json

@@ -1,7 +1,8 @@
 {
     "schema": "https://github.com/Project-MONAI/MONAI-extra-test-data/releases/download/0.8.1/meta_schema_20220324.json",
-    "version": "0.3.5",
+    "version": "0.3.6",
     "changelog": {
+        "0.3.6": "enhance readme with details of model training",
         "0.3.5": "update to use monai 1.0.1",
         "0.3.4": "enhance readme on commands example",
         "0.3.3": "fix license Copyright error",

docs/README.md

@@ -4,13 +4,31 @@ A pre-trained model for volumetric (3D) segmentation of the spleen from CT image
 # Model Overview
 This model is trained using the runner-up [1] awarded pipeline of the "Medical Segmentation Decathlon Challenge 2018" using the UNet architecture [2] with 32 training images and 9 validation images.
 
+![image](https://developer.download.nvidia.com/assets/Clara/Images/clara_pt_spleen_ct_segmentation_workflow.png)
+
 ## Data
 The training dataset is Task09_Spleen.tar from http://medicaldecathlon.com/.
 
 ## Training configuration
-The training was performed with at least 12GB-memory GPUs.
+The segmentation of spleen region is formulated as the voxel-wise binary classification. Each voxel is predicted as either foreground (spleen) or background. And the model is optimized with gradient descent method minimizing Dice + cross entropy loss between the predicted mask and ground truth segmentation.
+
+The training was performed with the following:
+
+- GPU: at least 12GB of GPU memory
+- Actual Model Input: 96 x 96 x 96
+- AMP: True
+- Optimizer: Adam
+- Learning Rate: 1e-4
+- Loss: DiceCELoss
 
-Actual Model Input: 96 x 96 x 96
+Pre-processing transforms:
+
+1. Convert data to channel-first
+2. Resample to resolution 1.5 x 1.5 x 2 mm
+3. Scale intensity
+4. Cropping foreground surrounding regions
+5. Cropping random fixed sized regions of size [96,96,96] with the center being a foreground or background voxel at ratio 1 : 1
+6. Randomly shifting intensity of the volume
 
 ## Input and output formats
 Input: 1 channel CT image
@@ -22,6 +40,17 @@ This model achieves the following Dice score on the validation data (our own spl
 
 Mean Dice = 0.96
 
+## Training Performance
+A graph showing the training loss over 1260 epochs (10080 iterations).
+
+![](https://developer.download.nvidia.com/assets/Clara/Images/clara_pt_spleen_ct_segmentation_train_2.png) <br>
+
+## Validation Performance
+A graph showing the validation mean Dice over 1260 epochs.
+
+![](https://developer.download.nvidia.com/assets/Clara/Images/clara_pt_spleen_ct_segmentation_val_2.png) <br>
+
+
 ## commands example
 Execute training: