0% found this document useful (0 votes)
13 views54 pages

Medical Image Preprocessing Techniques

The document discusses biomedical signal and image analysis, focusing on data preprocessing techniques essential for medical images. Key preprocessing steps include background removal, denoising, normalization, and enhancement to improve image quality and comparability. It also covers data augmentation methods and various segmentation techniques, such as Otsu's thresholding and region growing, to extract relevant information from medical images.

Uploaded by

b22210
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
13 views54 pages

Medical Image Preprocessing Techniques

The document discusses biomedical signal and image analysis, focusing on data preprocessing techniques essential for medical images. Key preprocessing steps include background removal, denoising, normalization, and enhancement to improve image quality and comparability. It also covers data augmentation methods and various segmentation techniques, such as Otsu's thresholding and region growing, to extract relevant information from medical images.

Uploaded by

b22210
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Biomedical Signal &

Image Analysis
Data Pre-Processing
Data Processing
● Medical Images cannot be utilized similarly it was procured.
● Image preprocessing prepares data for a target workflow.
● The main goals of medical image preprocessing are to reduce image acquisition
artifacts and to standardize images across a data set.
● Exact preprocessing requirements depend on :
○ The modality and procedure used to acquire data,
○ Target workflow.
● Common pre-processing steps are:
○ Background removal,
○ Denoising,
○ Resampling,
○ Registration,
○ Intensity/color normalization
○ Contrast Enhancement
Data Processing: Framework
Data Normalization

Process that adjusts the intensity values or spatial relationships within biomedical
images to improve consistency and comparability across different images
Data Normalization can be of two types:
1. Intensity Normalization
a. Min-Max
b. Z-Score (Standardization)
c. Histogram Matching
2. Spatial Normalization
a. Registration
b. Resampling
Intensity Normalization: Min-Max

Image normalization aims at correcting the heterogeneous intensity between cases.


Intensity Normalization: Min-Max
Intensity Normalization: Min-Max
Intensity Normalization: Z-Score
Intensity Normalization: Z-Score
Z-score normalization assumes a normal
distribution. Other techniques, like
Robust Scaler or Quantile
Transformation can be more
appropriate.

Subject A

Subject B

Subject C
Intensity Normalization: Z-Score
Normalization: Z-Test
Enhancement: Histogram
Enhancement: Histogram

?
Histogram: Stretching

Extending
the dynamic
s=T(r) range
Histogram: Stretching

Smax= L-1 , Smin = 0; L = 2bpp

Old Histogram New Histogram


Histogram: Equalization
Histogram Processing: Equalization
Histogram Processing: Equalization
Histogram Processing: Equalization
Histogram Equalization: CLAHE
Histogram: Matching

1 2
Histogram: Matching
3
Histogram: Matching
Spatial Normalization: Image Registration

Global Transformation Models


Spatial Normalization: Resampling

Many datasets do not have equal sampling rates across all


dimensions.

We can get the sampling rates along each dimension from


the metadata dictionary
Spatial Normalization: Resampling vs Resizing
Spatial Normalization: Registration
Noise Filtering
Noise: Types
Noise: Types
Gaussian Speckle

Impulse
Denoising: Techniques
Homomorphic Filters

An image f(x, y) can be expressed as the product of its illumination, i(x, y), and reflectance,
r(x, y) components:

However;

The illumination component of an image generally is characterized by slow spatial variations,


while the reflectance component tends to vary abruptly, particularly at the junctions of
dissimilar objects.
Homomorphic Filters

(a) Full body PET scan.


(b) Image enhancement using
homomorphic filtering.
Noise Filtering: Spatial/Frequency Domain
Color Normalization

To reduce variations in color due to factors like staining inconsistencies, lighting,


and scanner variations.

● Histogram Specification
● Reinhard Method
● Vahadane
● Generative Adversarial Networks
● Color Deconvolution method
● CLAHE etc
Color Normalization: Reinhard

Transformed Components
Source
Components

Reference
Components
Data Augmentation
● Data Augmentation takes existing medical images and applies various transformations to
create a plethora of varied yet related images from a limited dataset
● These Augmentation techniques are
○ Rotation, Translation
○ Rescaling
○ Flipping, and Shearing
○ Random Cropping and Erasing
○ Blurring
○ Mirroring
○ Brightness, Sharpness, Contrast Enhancement
○ Adding Noise
○ Color Modification
○ Synthetic features etc
Data Augmentation
First order Derivative: Gradient Filter

Definition we use for a first derivative:


1. must be zero in flat segments (areas of constant gray-level values);
2. must be nonzero at the onset of an intensity step or ramp;
3. must be nonzero along ramps.

Definition we use for second derivative:


1. must be zero in flat areas;
2. must be nonzero at the onset and end of an intensity step
3. must be zero along ramps of constant slope.
Second order Derivative: Laplacian Filter

The simplest isotropic derivative operator is the Laplacian, which, for a function f(x, y) of two
variables, is defined as

Because derivatives of any order are linear operations, the Laplacian is a linear operator.

*c is negative if the center pixel of operator mask


Image Sharpening:
is negative
First order Derivative: Gradient Filter
First order Derivative: Gradient Filter

Roberts cross-gradient Sobel


Effect of Noise
Detection : Derivative of Gaussian
LOG: Laplacian of Gaussian
ROI Extraction: Segmentation

NOW

● Graph based
○ Graph cut
● Supervised ML
○ RF
○ SVM
○ kNN
○ Bayesian Net etc
● Deep Learning
○ CNN
○ RNN
○ GAN
○ LSTM
○ YOLO
○ U-Net
○ DBN etc
Segmentation: Thresholding
T

Single threshold value for entire image/Intensity Histogram

Automatic Thresholding
Segmentation: Otsu’s Thresholding

Optimum Global Thresholding: Statistical Decision

Otsu’s method
minimizes the overall within-class variance
OR
maximizing between class variance by
minimizing the weighted sum of class
variances
Segmentation: Otsu’s Thresholding

1. Compute the histogram of the image.


2. Let each gray level have probability pi.
3. Compute the cumulative sum P1(k) for k=0,…,L-1

4. Compute the cumulative mean m(k) for k=0,…,L-1

5. Compute the global intensity mean mG ;

6. Compute the between-class variance for k=0,…,L-1


Separability Measure

7. Compute the value of k for which 𝝈2B(k) is maximum

Otsu’s threshold
Segmentation: Otsu’s Thresholding
Segmentation: Otsu’s Thresholding
Segmentation: Region Growing

● A region can be classified as a group of connected pixels exhibiting similar properties.


● The “similarity” or “connected pixels” can be identified in terms of:
○ Intensity
○ Color
○ Motion
○ Texture
○ Shape
○ Topology, etc.

● In this type of segmentation, some predefined rules are present which have to be obeyed by a pixel
in order to be classified into similar pixel regions.
● Region-based segmentation methods are preferred over edge-based segmentation methods in
case of a noisy image.
● Region-Based techniques are further classified into 2 types based on the approaches they follow:
○ Region Growing method
○ Region Splitting and Merging method
Segmentation: Region Growing

Selection of Seed Points:


● To start from some pixels (seed points, S(x,y)) representing distinct image regions

● If no prior information available, seed points calculated from the centroid of the clusters
or group of pixels.
Similarity Criteria:
predicate/seeds within Ri
● Depends on problems/types of images/spatial proximity
● Growth mechanism and Homogeneity checking
P(R) : abs{I(x,y)-S(x,y)} ≤ Threshold

Formulation of Stopping Rule


Segmentation: Region Growing

Threshold = 1

You might also like