# hybrid-cs-model-mri¶

## Use Case and High-Level Description¶

The hybrid-cs-model-mri model is a hybrid frequency-domain/image-domain deep network for Magnetic Resonance Image (MRI) reconstruction. The model is composed of a k-space network that essentially tries to fill missing k-space samples, an Inverse Discrete Fourier Transformation (IDFT) operation, and an image-domain network that acts as an anti-aliasing filter.

More details provided in the paper and repository.

## Specification¶

Metric

Value

Type

MRI Image Inpainting in k-Space

GFlops

146.6037

MParams

11.3313

Source framework

TensorFlow*

## Accuracy¶

Accuracy metrics are obtained on Calgary-Campinas Public Brain MR Dataset.

Metric

Value

PSNR (mean)

34.272884 dB

PSNR (std)

4.607115 dB

Use accuracy_check [...] --model_attributes <path_to_folder_with_downloaded_models> to specify the path to additional model attributes. path_to_folder_with_downloaded_models is a path to the folder, where models are downloaded by Model Downloader tool.

## Input¶

### Original model¶

MRI input, name - input_1, shape - 1, 256, 256, 2, format - B, H, W, C, where:

• B - batch size

• H - image height

• W - image width

• C - number of channels

### Converted model¶

MRI input, name - input_1, shape - 1, 256, 256, 2, format - B, H, W, C, where:

• B - batch size

• H - image height

• W - image width

• C - number of channels

## Output¶

### Original model¶

The net outputs a blob with the name StatefulPartitionedCall/model/conv2d_43/BiasAdd/Add and shape 1, 1, 256, 256, containing reconstructed MR image.

### Converted model¶

The net outputs a blob with the name StatefulPartitionedCall/model/conv2d_43/BiasAdd/Add and shape 1, 1, 256, 256, containing reconstructed MR image.

You can download models and if necessary convert them into OpenVINO™ IR format using the Model Downloader and other automation tools as shown in the examples below.

omz_downloader --name <model_name>
omz_converter --name <model_name>