Biomedical Engineering:

Analysis of Biomedical Data and

Image for Diagnosis

Course Instructor :
Prof. Mahua Bhattacharya

ABV Indian Institute of Information Technology &

Management,

Gwalior 474010, Madhya Pradesh

1
 Medical Image/data
Processing for Different
Biomedical Applications for
diagnostic planning using
computational approaches
 What are the different aspects / categories
of medical data analysis research
Collection of raw medical data
Processing of data
Segmentation of ROI (region of interest) as per
medical observation
Identification / classification of lesion for surgical
assistance
Gradation /identification of different stages of
prognosis
Medical image registration for different modality
images
Information combination using fusion techniques
 Need of Multimodality Image / Data Acquisition and
Analysis
During the last few decades there is remarkable advancement in
diagnostic radiology and nuclear imaging techniques

Different diagnostic modalities like X-rays, Computed

Tomography (CT), Magnetic Resonance Imaging (MRI),
Positron Emission Tomography (PET), Single Photon
Emission Computed Tomography (SPECT),
Ultrasonography (USG) are increasingly being used by the
medical practitioners for diagnosing diseases and planning
treatment procedures

Different innovative image processing and analysis

techniques are being adopted to these modalities to get more and
more information about internal structures of human
organs.
Some Applications On Medical Image /
Data Analysis
 Other Applications of BME in Cell Level and
Molecular Levels

The automatic cell analysis method is capable of segmenting the cells and
can detect the number of live/dead cells present in the body.
This study proposed a novel non-linear segmentation model (NSM) for the
segmentation and quantification of live/dead cells present in the body. This
work also reveals the aspects of electromagnetic radiation on the cell body.
The bright images of the hippocampal CA3 region of the rat brain under
the resolution of 60× objective are used to analyze the effects called NISSL-
stained dataset. The proposed non-linear segmentation model segments the
foreground cells from the cell images based on the linear regression
analysis. These foreground cells further get discriminated as live/dead cells
and quantified using shape descriptors and geometric method, respectively.
The counting analysis of live and dead cells using the proposed method is
far better than the manual counts. The segmentation model and
quantifying procedure is an amalgamated method for cell quantification
that yields better segmentation results 6
Key Features ( Novelty or Specialty)
 Medical
science

Statistics
Mathematics

Medical
AI and
Image /data
Computer Softcomputing
analysis: An
Vision Interdisciplinary
Research
Communication

Basic and
Applied Science
Management
Science
 Approaches For Medical Data analysis
For Analysis of medical Image Segmentation is a
crucial step for detection the Region of Interest

(a) Application of segmentation techniques on MR

brain images of CT and MR modalities

(b) Segmentation techniques on mammographie

images for breast cancer screening
 Image Pre-Processing
Image Preprocessing is a technique to process digital image to
remove the unwanted information from the image and to prepare the
image for next phase of processing.

This unwanted information may appear in the image in the form of

noises of different characteristics and which are required to be
removed to extract the useful information.

Image Preprocessing is also called the process of

“Image Enhancement” for better visibility of the image.

Contrast Enhancement is also part of image preprocessing

Image Pre processing is not generating any new information in the

image it is enhancing the existing information of the image for its
analysis and interpretation
 Image Enhancement in Spatial Domain

Spatial Domain approach on digital image is represented as :

g( x,y) = T f (x,y) , where f(x,y) is the input image and g(x,y) is the
processed output image and T is the operator

In digital image processing the basic mathematical operation which is

used is 2D mathematical convolution between two functions which are
expressed as two matrices for Processing

Spatial Filtering Fundamentals

To remove the noise from the digital image which is represented by an

matrix array a small subimage matrix of dimension (3 by 3 or 5 by 5 ,
or 7 by 7 etc.) which is called a mask, template , kernel or window is
designed . The values of the subimage matrix is called
“co efficient” or “element” 12
Smoothing Spatial Filter

Smoothing filter is also called as Average Filter or Low Pass Filter.

Smoothing filters are used to remove Random Gaussian Noise and the
effect of the filter on image is Blurring.

The mask or the subimage matrix is having mask element value 1.

The idea how a Smoothing Filter works - by replacing the value of
every pixel in an image by the average of the grey levels in the
neighborhood defined by the subimage

13
 Averaging is a special case of discrete convolution. For a 3 x 3
neighborhood the convolution mask h is

A Convolution Mask

The above equation is equivalent to discrete convolution with the

kernel h, that is called a convolution mask.
Median Statistical Spatial Filter

Median filter is used to remove Impulsive Noise from the digital

image which is also called as Salt-and-Pepper Noise:

Median filter is based on the ordering / sorting process of the pixel

contained in an 3 X 3 , 5 X 5 area and then to replace the value of the
center pixel with the value of the sorting result - to replace the pixels by
the median of the grey levels in the neighborhood of the pixels.

For 3 X 3 area the 5th largest value in 5 X 5 the 12th largest value of
grey level for the middle of the

15
Image corrupted by Impulse noise or Salt and Pepper Noise

16
Sharpening Spatial Filter

This filter is useful to restore the finer details and to restore the
blurring effect.
The sharpening filter is based on derivative of the image
function, ( gradient function)
The derivative operators are proportional to the degree of

discontinuity at the point where the operator is applied

Image differentiation enhances edges and other finer details

The first order derivative of 1D function f(x) as:

For image function of f(x,y), the partial derivatives are used

where f has variables x and y
Intensity Based Preprocessing :
Images are required to be enhanced when due to poor illumination and
lack of dynamic range of the imaging sensor images are having low visibility

The intensity distribution / grey level distribution in the mage is not uniform
and which creates poor visibility of the image . The most common process is
Contrast Stretching - the idea is to stretch the grey levels over the entir
dynamic range of the image

A Low Contrast Image Result of Contrast Streching

18
 Image Segmentation
Image segmentation in digital image processing is a process to subdivide
an image into its constituent components / regions/ objects to isolate
the Region of Interest (ROI) from the entire Image

Segmentation is based on two basic properties of Intensity values

Discontinuity - to partition an image based on sharp or abrupt change in
intensity (Such as Edge )

Similarity : To partition an image into regions of similar property with

predefined criteria
Detection of Discontinuities

To detect the three basic types of gray-level discontinuities -

points, lines, edges, the common way is to run a mask through the
image
 Point Detection

Mask for point Detection

A point has been detected at the location on which the mark is centered if
|R| ≥ T
where T is a nonnegative threshold R is the sum of products of the
coefficients with the gray levels contained in the region encompassed
by the mark.
Edge Detection Or Edge Segmentation

Edge in an digital image is defined as the digital boundary between the

background and the object region where the abrupt intensity changes are
occurring w.r.t. the neighboring regions

Edge detection is an important step for segmentation to identify the object

of interest from the image

Isolated edge points are detected by the process of edge segmentation

finally to form a closed boundary using an edge linking algorithm

Edge detectors are designed based on derivatives of image function

f (x,y) which are specified as Derivative or Gradient Operator

First-order derivative (Gradient operator

Second-order derivative (Laplacian operator)
22
Gradient Operator

•first order derivatives are implemented using the magnitude of the gradient.

commonly approx.

the magnitude becomes nonlinear

83
First Order Gradient Masks

84
Diagonal Edges with Prewitt and Sobel
Masks

85
Laplacian Operator
which is for an image function f(x,y) of two variables is defined as:

Partial Second Order Derivatives in the X direction

Partial Second Order Derivatives in the y direction :

26
Two Dimensional Laplacian Equation is obtained by adding the two
expressions

Laplacian Operator : Second Order

Derivative for Edge Detection

27
First Order Gradient Masks

84
Diagonal Edges with Prewitt and Sobel
Masks

85
 Canny Advanced Edge Detection Technique

The Canny Edge Detector is an edge detection operator that uses

a multi-stage algorithm to detect a wide range of edges in images.
It was developed by John F. Canny in 1986. Canny also
produced a computational theory of edge detection explaining
why the technique works.

Canny Edge detection has suggested Edge Linking Algorithm

for a complete contour of the object of recognition
Region Based Segmentation Method

The region based segmentation methods are the methods that

segments the image into various regions having similar
characteristics. There are two basic techniques based
on this method:

1. Region growing methods

2. Region splitting and merging methods

Locate tumors and other pathologies
Measure tissue volumes
Study of anatomical structure

Diagnosis, Treatment planning, Computer-

guided surgery
segmentation refers to the process of partitioning a
digital image into multiple segments
The goal of segmentation is to simplify and/or change
the representation of an image into something that is
more meaningful and easier to analyze
Image segmentation is typically used to locate objects
and boundaries (lines, curves, etc.) in images
Segmentation of brain MR images into three
different cortical tissues :-

Gray matter
white matter
cerebrospinal fluid

The volumetric changes in these brain tissues help

in the study of neural disorders
Some Process of Segmentation for Brain MR
data
1 Fuzzy C-means
2 Markov random field model

Evaluation methods for image segmentation are classified

into analytical and empirical evaluation methods

The analytical methods focus on analyzing the properties

of a segmentation algorithm, such as its processing
strategy, complexity, and efficiency
Empirical methods are further classified into
1 Goodness methods,
2 Discrepancy methods
The primary difficulty in achieving an
accurate segmentation is the intensity in-
homogeneities which commonly occur in MR
image
FCM clustering is a data clustering algorithm in
which each data point belongs to a cluster with a
degree specified by a membership grade
The data samples may belong to more than one
group with a varying membership value ranging from
0 to 1
Major advantage of FCM over K-means clustering.
Formation of new clusters in FCM is possible by
monitoring data points having close membership
values to the existing classes
 Computational Basis of Fuzzy C Mean
Let U = {u1, u2, . . . , un} be a set of given data. A fuzzy c-partition of
U is a family of
fuzzy subsets of U., denoted by P = {A1, A2, . . . ,Ac}, which satisfies

The performance index of a fuzzy partition P, Im(P), is defined in

terms of cluster centers by the formula

Where vi is the cluster center for ith cluster and ||uk −vi ||
represents the distance d, between uk and vi . Clearly, the
smaller the value of Im(P), the better the fuzzy partition P.
Thus, fuzzy partition has a goal to minimize the performance
index Im(P), which offers
The FCM algorithm is an iterative procedure as described below:
1. Initialise the number of classes and the membership matrix Ai (uk )
with random values between 0 and 1 such that the Eq. (1) is satisfied.
2. Calculate the fuzzy clusters centers vi , i = 1, 2, . . . , c, using the Eq.
(3).
3. Iterate until the improvement over the previous iteration is below a
certain threshold, this is done by computing the cost function using the
Eq. (2).
4. Compute the new membership matrix Ai (uk ) using Eq. (4). Go to
step-2.

Thus, by iteratively updating the cluster centers and the membership

degrees for each data point, the FCM algorithm iteratively moves
towards a local minimum.
Segmentation methods based on Markov random field (MRF)
model provide a convenient way to combine both the
conditional intensity distribution of pixel intensities and of
contextual information

The conditional intensity says that pixel intensities of nearly

homogeneous region will follow a certain statistical distribution
and

Contextual information is based on the property that pixels

which are lying in the neighbourhood will tend to have similar
intensity values

The task here is to find out the true label of each pixel or voxel
which may belong to the subset of {GM, WM, CSF}.
Intensity distribution and Contextual
information are combined using MRF models
to obtain a powerful decision rule regarding the
true label of the pixel

MRF is defined by the following two properties

Positive definiteness
Markov property

Any MRF segmentation algorithm includes three main

properties—
neighbourhood correlations
non-parametric statistics and
signal in-homogeneities
The accuracy is the proportion of total number
of pixels that were segmented correctly

NP is the number pixels segmented as positive

when the actual is negative,
PN is the number of pixels which have been
segmented as negative when the actual is
positive and
PP is the number of pixels when both
segmented and actual are positive
The true positive rate (TP) is the proportion of number
of pixels predicted as they belong to a particular region
R and are segmented correctly

The false positive rate (FP) is the proportion of number

of pixels which are predicted as they do not belong to
region R and are incorrectly segmented that they belong
to region R
Figure 2: (a) MR image slice obtained from BrainWeb Simulated
Database (b) Ground truth showing WM in yellow, CSF in blue, GM in
green
Basic approach using intensities based on histogram
Disregards the spatial context within which the
intensity value occurs
Intensively used for the initial segmentation of images prior
to more sophisticated segmentation methods for the
purpose of reduction of convergence time

Figure 3: Initial labeling based on thresholding

Segmentation results. a FCM result b ICM result c Metropolis result d Gibbs sampling resu

Histograms of segmented images. a FCM b ICM c Metropolis d Gibbs sampling

PROCESS SEGMENTATION FOLLOWED

T1 weighted brain MR image taken from Brain

Web simulated database is used for segmentation
and evaluation

The number of regions is set to 3 and the skull

stripped image is segmented using fuzzy c-means
clustering

For segmenting the image with iterated

conditional modes, the image is initially segmented
using the thresholding method

Both ICM and gibbs sampling are done for 5

iterations each
CSF,WM and GM of ground truth and segmented images. a Original image bFCMc ICM d Metropolis e Gibbs
sampling
Discrepancy methods
Accuracy:

All the three methods are able to segment the CSF almost
correctly leaving few corners
In case of FCM, there is a lack of continuity for
segmenting GM and WM
It is seen that WM is mistaken for GM
This is due to not considering contextual information in
FCM and the difference between the intensities of GM and
WM is very less
In case of FCM, there is a lack of continuity for
segmenting GM and WM
The accuracy for segmenting WM using FCM is only
39% since FCM is not considering contextual
information
The approaches which use Markov random field
model have been able to successfully segment
WM and GM

Gibbs sampling and Metropolis algorithms give a better

solution for the optimization problem