diff --git a/models/brats_mri_generative_diffusion/docs/README.md b/models/brats_mri_generative_diffusion/docs/README.md
index 1c01d861..444db62e 100644
--- a/models/brats_mri_generative_diffusion/docs/README.md
+++ b/models/brats_mri_generative_diffusion/docs/README.md
@@ -1,11 +1,11 @@
 # Model Overview
 A pre-trained model for volumetric (3D) Brats MRI 3D Latent Diffusion Generative Model.
 
-This model is trained on BraTS 2016 and 2017 data from [Medical Decathlon](http://medicaldecathlon.com/), using the Latent diffusion model [1].
+This model is trained based on BraTS 2018 data from [Multimodal Brain Tumor Segmentation Challenge (BraTS) 2018](https://www.med.upenn.edu/sbia/brats2018.html), using the Latent diffusion model [1].
 
 ![model workflow](https://developer.download.nvidia.com/assets/Clara/Images/monai_brain_image_gen_ldm3d_network.png)
 
-This model is a generator for creating images like the Flair MRIs based on BraTS 2016 and 2017 data. It was trained as a 3d latent diffusion model and accepts Gaussian random noise as inputs to produce an image output. The `train_autoencoder.json` file describes the training process of the variational autoencoder with GAN loss. The `train_diffusion.json` file describes the training process of the 3D latent diffusion model.
+This model is a generator for creating images like the T1CE MRIs based on BraTS 2018 data. It was trained as a 3d latent diffusion model and accepts Gaussian random noise as inputs to produce an image output. The `train_autoencoder.json` file describes the training process of the variational autoencoder with GAN loss. The `train_diffusion.json` file describes the training process of the 3D latent diffusion model.
 
 In this bundle, the autoencoder uses perceptual loss, which is based on ResNet50 with pre-trained weights (the network is frozen and will not be trained in the bundle). In default, the `pretrained` parameter is specified as `False` in `train_autoencoder.json`. To ensure correct training, changing the default settings is necessary. There are two ways to utilize pretrained weights:
 1. if set `pretrained` to `True`, ImageNet pretrained weights from [torchvision](https://pytorch.org/vision/stable/_modules/torchvision/models/resnet.html#ResNet50_Weights) will be used. However, the weights are for non-commercial use only.
@@ -20,12 +20,21 @@ An example result from inference is shown below:
 **This is a demonstration network meant to just show the training process for this sort of network with MONAI. To achieve better performance, users need to use larger dataset like [Brats 2021](https://www.synapse.org/#!Synapse:syn25829067/wiki/610865) and have GPU with memory larger than 32G to enable larger networks and attention layers.**
 
 ## Data
-The training data is BraTS 2016 and 2017 from the Medical Segmentation Decathalon. Users can find more details on the dataset (`Task01_BrainTumour`) at http://medicaldecathlon.com/.
+The training data is from the [Multimodal Brain Tumor Segmentation Challenge (BraTS) 2018](https://www.med.upenn.edu/sbia/brats2018.html).
 
 - Target: Image Generation
 - Task: Synthesis
 - Modality: MRI
-- Size: 388 3D volumes (1 channel used)
+- Size: 285 3D volumes (1 channel used)
+
+The provided labelled data was partitioned, based on our own split, into training (200 studies), validation (42 studies) and testing (43 studies) datasets.
+
+### Preprocessing
+The data list/split can be created with the script `scripts/prepare_datalist.py`.
+
+```
+python scripts/prepare_datalist.py --path your-brats18-dataset-path
+```
 
 ## Training Configuration
 If you have a GPU with less than 32G of memory, you may need to decrease the batch size when training. To do so, modify the `train_batch_size` parameter in the [configs/train_autoencoder.json](../configs/train_autoencoder.json) and [configs/train_diffusion.json](../configs/train_diffusion.json) configuration files.
@@ -34,46 +43,42 @@ If you have a GPU with less than 32G of memory, you may need to decrease the bat
 The autoencoder was trained using the following configuration:
 
 - GPU: at least 32GB GPU memory
-- Actual Model Input: 112 x 128 x 80
+- Actual Model Input: 128 x 128 x 128
 - AMP: False
 - Optimizer: Adam
-- Learning Rate: 1e-5
+- Learning Rate: 2e-5
 - Loss: L1 loss, perceptual loss, KL divergence loss, adversarial loss, GAN BCE loss
 
 #### Input
-1 channel 3D MRI Flair patches
+1 channel 3D MRI T1CE patches
 
 #### Output
 - 1 channel 3D MRI reconstructed patches
-- 8 channel mean of latent features
-- 8 channel standard deviation of latent features
+- 4 channel mean of latent features
+- 4 channel standard deviation of latent features
 
 ### Training Configuration of Diffusion Model
 The latent diffusion model was trained using the following configuration:
 
 - GPU: at least 32GB GPU memory
-- Actual Model Input: 36 x 44 x 28
+- Actual Model Input: 48 x 48 x 32
 - AMP: False
 - Optimizer: Adam
-- Learning Rate: 1e-5
+- Learning Rate: 2e-5
 - Loss: MSE loss
 
 #### Training Input
-- 8 channel noisy latent features
+- 4 channel noisy latent features
 - a long int that indicates the time step
 
 #### Training Output
-8 channel predicted added noise
+4 channel predicted added noise
 
 #### Inference Input
-8 channel noise
+4 channel noise
 
 #### Inference Output
-8 channel denoised latent features
-
-### Memory Consumption Warning
-
-If you face memory issues with data loading, you can lower the caching rate `cache_rate` in the configurations within range [0, 1] to minimize the System RAM requirements.
+4 channel denoised latent features
 
 ## Performance