PreprocessForNNUNetInference

MacroModule

genre	ML_Inference
author	Jan Hendrik Moltz
package	FMEstable/ReleaseMeVis
definition	PreprocessForNNUNetInference.def
keywords	deep, learning, nnUNet, convert, create, generate, input, RedLeaf

Purpose

This module applies a similar preprocessing as nnU-Net and was developed as part of the NNUNetInference macro that allows to run inference on models trained with nnU-Net in MeVisLab. If you use it outside NNUNetInference, you should be familiar with nnU-Net and know what you are doing, but that is typically not necessary.

Usage

The parameters need to be set up according to the plans.pkl or plans.json file that nnU-Net created when training your model. Connect your image and press Update.

Details

The module performs the preprocessing steps that nnU-Net’s own inference would also perform (see http://doi.org/10.1038/s41592-020-01008-z for details):

1. Non-zero cropping: This is always done by nnU-Net but can be disabled by setting Crop input to non-zero regions because it typically has no effect other than changing the output size.

2. Resampling: nnU-Net expects a fixed input voxel size which is set by Voxel Size. By default nnU-Net uses third-order spline resampling which is here replaced by Lanczos3 resampling for ease of implementation. Like nnU-Net, we use nearest neighbor resampling for the out-of-plane dimension on anisotropic images (maxVoxelSize / minVoxelSize > 3).

3. Intensity normalization: nnU-Net has different normalization schemes depending on the image modality. Normalization Scheme “none” does not normalize at all, “nonCT” subtracts mean and standard deviation of inImage (aka z-score normalization), “CT” uses fixed Mean and Standard Deviation (assuming Hounsfield units) and additionally clips at Lower Clipping Threshold and Upper Clipping Threshold, and “CT2” uses only clipping and no z-scoring. It is also possible to Use only non-zero regions for nonCT and CT2 z-score, which happens in nnU-Net if non-zero cropping reduces the image size by more than 25%.

4. Padding: If after cropping and resampling the image is smaller than the desired patch size (Minimum Output Image Size), it is zero-padded.

Windows

Default Panel

Input Fields

inImage

name: inImage, type: Image, deprecated name: input0

Image to be preprocessed (and then to be fed into a trained nnU-Net model)

Output Fields

outImage

name: outImage, type: Image, deprecated name: output0

Preprocessed image that can be fed into a trained nnU-Net model.

outImageWithoutPadding

name: outImageWithoutPadding, type: Image, deprecated name: output0WithoutPadding

Same image as outImage but without padding, i.e. potentially smaller than desired patch size.

Parameter Fields

Field Index

[]: Trigger	Normalization Scheme: Enum	Voxel Size: Vector3
Clear: Trigger	On Input Change Behavior: Enum
Crop input to non-zero regions: Bool	Standard Deviation: Double
doNotClearOnFailedUpdate: Bool	Status Code: Enum
Has Valid Output: Bool	Status Message: String
Lower Clipping Threshold: Double	Update: Trigger
Mean: Double	Upper Clipping Threshold: Double
Minimum Output Image Size: IntVector3	Use only non-zero regions for nonCT and CT2 z-score: Bool

Visible Fields

Update

name: update, type: Trigger

Initiates update of all output field values.

Clear

name: clear, type: Trigger

Clears all output field values to a clean initial state.

On Input Change Behavior

name: onInputChangeBehavior, type: Enum, default: Clear, deprecated name: shouldAutoUpdate,shouldUpdateAutomatically

Declares how the module should react if a value of an input field changes.

Values:

Title	Name	Deprecated Name
​Update	​Update	​TRUE
​Clear	​Clear	​FALSE

[]

name: updateDone, type: Trigger, persistent: no

Notifies that an update was performed (Check status interface fields to identify success or failure).

Has Valid Output

name: hasValidOutput, type: Bool, persistent: no

Indicates validity of output field values (success of computation).

Status Code

name: statusCode, type: Enum, persistent: no

Reflects module’s status (successful or failed computations) as one of some predefined enumeration values.

Values:

Title	Name
​Ok	​Ok
​Invalid input object	​Invalid input object
​Invalid input parameter	​Invalid input parameter
​Internal error	​Internal error

Status Message

name: statusMessage, type: String, persistent: no

Gives additional, detailed information about status code as human-readable message.

Crop input to non-zero regions

name: inCropNonZeroRegions, type: Bool, default: TRUE

If true, the inImage is reduced to the bounding box of all non-zero voxels (after ApplyGlobalModalityLUT).

Use only non-zero regions for nonCT and CT2 z-score

name: inUseNonZeroRegionForZScore, type: Bool, default: FALSE

If true, only non-zero voxels (AFTER normalizing image values regarding DICOM Rescale Intercept and Slope) are considered when computing mean and standard deviation for z scoring (in nonCT and CT2 z-score normalization mode).

Minimum Output Image Size

name: inMinimumOutImageSize, type: IntVector3, default: 64 64 64, deprecated name: inPatchSize

Desired minimum output image size. Should be equals to the patch size of the neural network applied later on.

Voxel Size

name: inVoxelSize, type: Vector3, default: 1 1 1

Desired output voxel size.

Normalization Scheme

name: inNormalizationScheme, type: Enum, default: nonCT

Scheme by which image values are normalized.

Values:

Title	Name	Description
​CT	​CT	​Normalize by z-scoring with Mean and Standard Deviation and clipping with Lower Clipping Threshold and Upper Clipping Threshold.
​CT2	​CT2	​Normalize by clipping with Lower Clipping Threshold and Upper Clipping Threshold.
​nonCT	​nonCT	​Normalize by z-scoring with mean and standard deviation of input.
​none	​none	​Do not normalize.

Lower Clipping Threshold

name: inLowerClippingThreshold, type: Double, default: 0

Threshold below which image values are clipped if Normalization Scheme is CT or CT2.

Upper Clipping Threshold

name: inUpperClippingThreshold, type: Double, default: 0

Threshold above which image values are clipped if Normalization Scheme is CT or CT2.

Mean

name: inMean, type: Double, default: 0

Value that is subtracted from the image if Normalization Scheme is CT.

Standard Deviation

name: inStandardDeviation, type: Double, default: 0

Value by which the image is divided if Normalization Scheme is CT.

Hidden Fields

doNotClearOnFailedUpdate

name: doNotClearOnFailedUpdate, type: Bool, persistent: no

Prevents automated clear after update failed. This does not affect status fields. It enables the developer to analyze module’s state after failure.