Preprocessing and Augmentation

3-D registration and denoising, random intensity augmentation

Image preprocessing and image augmentation prepare data for advanced medical image analysis workflows. Use image preprocessing to reduce image acquisition artifacts and format data for the target workflow. For example, you can remove noise, normalize intensity values, resize image voxels, or align images using registration. Use image augmentation to increase the amount and variability of training data for deep learning workflows. For example, you can randomly adjust image contrast or apply random rotations or scaling to simulate variations in image acquisition and patient anatomy. To get started, see Get Started with Image Preprocessing and Augmentation for Deep Learning.

Functions

expand all

Denoising

`specklefilt`	Filter image using speckle-reducing anisotropic diffusion
`imfilter`	N-D filtering of multidimensional images
`medfilt2`	2-D median filtering
`medfilt3`	3-D median filtering
`imgaussfilt`	2-D Gaussian filtering of images
`imgaussfilt3`	3-D Gaussian filtering of 3-D images
`fspecial`	Create predefined 2-D filter
`fspecial3`	Create predefined 3-D filter

Registration

Fast Registration

imregmoment Fast registration of grayscale images or intensity volumes using moment of mass method

Optimization-Based Registration

`imregister`	Intensity-based image registration
`imregtform`	Estimate geometric transformation that aligns two 2-D or 3-D images
`imregconfig`	Configurations for intensity-based registration
`MattesMutualInformation`	Mattes mutual information metric configuration
`MeanSquares`	Mean square error metric configuration
`RegularStepGradientDescent`	Regular step gradient descent optimizer configuration
`OnePlusOneEvolutionary`	One-plus-one evolutionary optimizer configuration

Deformable Registration

`imregdeform`	Deformable registration of grayscale images or intensity volumes using total variation method
`imregdemons`	Estimate displacement field that aligns two 2-D or 3-D images

Groupwise Registration

imreggroupwise Groupwise deformable registration

Correlation-Based Registration

`imregcorr`	Estimate geometric transformation that aligns two 2-D images using phase correlation
`normxcorr2`	Normalized 2-D cross-correlation

Surface Registration

imregicp Surface registration using iterative closest point algorithm

Apply Transformation

`imwarp`	Apply geometric transformation to image
`resample`	Resample medical image volume in different patient coordinate system

Augmentation

`jitterIntensity`	Randomly augment intensity of grayscale image or intensity volume
`randomWindow2d`	Randomly select rectangular region in image
`randomCropWindow3d`	Create randomized cuboidal cropping window
`centerCropWindow2d`	Create rectangular center cropping window
`centerCropWindow3d`	Create cuboidal center cropping window
`Rectangle`	Spatial extents of 2-D rectangular region
`Cuboid`	Spatial extents of 3-D cuboidal region
`randomAffine2d`	Create randomized 2-D affine transformation
`randomAffine3d`	Create randomized 3-D affine transformation
`affineOutputView`	Create output view for warping images
`imerase`	Remove image pixels within rectangular region of interest

Topics

Medical Image Preprocessing
Learn common preprocessing steps used in medical image analysis workflows.
Medical Image Registration
Techniques for alignment of medical images, volumes, and surfaces.
Get Started with Image Preprocessing and Augmentation for Deep Learning
Preprocess data with deterministic operations such as normalization or color space conversion, or augment training data with randomized operations such as random cropping or color jitter.
Train and Apply Denoising Neural Networks
Use a pretrained neural network to remove Gaussian noise from a grayscale image, or train your own network using predefined layers.

Related Information

Image Filtering and Enhancement

Featured Examples

Register Multimodal Medical Image Volumes with Spatial Referencing

Align two medical image volumes using moment-of-mass-based registration.

Open Live Script

Unsupervised Medical Image Denoising Using CycleGAN

Generate high-quality high-dose computed tomography (CT) images from noisy low-dose CT images using a CycleGAN neural network.

Unsupervised Medical Image Denoising Using UNIT

Generate high-quality computed tomography (CT) images from noisy low-dose CT images using a UNIT neural network.