1 of 127

1

1/15/25

Lectures 3 & 4

Medical Image Processing - Basics

BME 495:

Deep Learning for Medical Imaging

Ulas Bagci, Ph.D.,

Director of Machine & Hybrid Intelligence Lab,�Northwestern University, Chicago

Machine & Hybrid Intelligence Lab

2 of 127

Outline of Lectures 3 & 4

Image Pre-Processing

Basics & Definitions

Filters

Machine & Hybrid Intelligence Lab

3 of 127

Some Useful Definitions Throughout This Course

3

Machine & Hybrid Intelligence Lab

4 of 127

4

CAD: Computer Aided Diagnosis

Science of underlying computerized methods for the diagnosis of diseases via images

Machine & Hybrid Intelligence Lab

5 of 127

Pre (Image) Processing

For enhancing information about and defining object system.

Visualization

For viewing and comprehending object system

Manipulation

For altering object system (virtual surgery)

Analysis

For quantifying information about object system.

5

CAD: Computer Aided Diagnosis

Science of underlying computerized methods for the diagnosis of diseases via images

Machine & Hybrid Intelligence Lab

6 of 127

Pre (Image) Processing

For enhancing information about and defining object system.

Visualization

For viewing and comprehending object system

Manipulation

For altering object system (virtual surgery)

Analysis

For quantifying information about object system.

6

CAD: Computer Aided Diagnosis

Science of underlying computerized methods for the diagnosis of diseases via images

Machine & Hybrid Intelligence Lab

7 of 127

Pre (Image) Processing

For enhancing information about and defining object system.

Visualization

For viewing and comprehending object system

Manipulation

For altering object system (virtual surgery)

Analysis

For quantifying information about object system.

7

CAD: Computer Aided Diagnosis

Science of underlying computerized methods for the diagnosis of diseases via images

Volume of Interest (VOI):

Purpose: to reduce data for speeding up CAD operations.

Region of Interest (ROI):

To specify a subset of the scene domain.

Intensity of Interest (IOI):

To specify a subset of the scene intensities.

Storage requirement can be reduced by a factor of 2-10

Machine & Hybrid Intelligence Lab

8 of 127

VOI/ROI (Volume/Region of Interest)

8

Machine & Hybrid Intelligence Lab

9 of 127

VOI/ROI (Volume/Region of Interest)

9

CUT

Machine & Hybrid Intelligence Lab

10 of 127

VOI/ROI (Volume/Region of Interest)

10

Kidney region

Machine & Hybrid Intelligence Lab

11 of 127

VOI/ROI (Volume/Region of Interest)

11

Machine & Hybrid Intelligence Lab

12 of 127

Spatial Resolution

12

Determines the smallest structure that can be represented in a digital image

(DPI: dots per inch)

Machine & Hybrid Intelligence Lab

13 of 127

13

Determines the smallest structure that can be represented in a digital image (DPI: dots per inch)

Spatial Resolution

Machine & Hybrid Intelligence Lab

14 of 127

14

Machine & Hybrid Intelligence Lab

15 of 127

Contrast Resolution

15

Local change in brightness and defined as the ratio between average brightness of an object and background

Machine & Hybrid Intelligence Lab

16 of 127

Radiographic Contrast

16

C: Center,

W: Width

Machine & Hybrid Intelligence Lab

17 of 127

Image Histogram ?

17

Machine & Hybrid Intelligence Lab

18 of 127

18

Image Histogram

It plots the number of pixels for each intensity value.

Machine & Hybrid Intelligence Lab

19 of 127

19

Image Histogram – Example Reading

Machine & Hybrid Intelligence Lab

20 of 127

20

Image Histogram – Example

Machine & Hybrid Intelligence Lab

21 of 127

Histogram based analysis of Lung CT (credit: Imbio)

21

Machine & Hybrid Intelligence Lab

22 of 127

Pixel Transformation

A function that takes an image (or images) an input and produces an output image (or scalar)

22

Machine & Hybrid Intelligence Lab

23 of 127

Pixel Transformation

A function that takes an image (or images) an input and produces an output image (or scalar)

23

X

Machine & Hybrid Intelligence Lab

24 of 127

Image Filtering

24

Purpose: To suppress unwanted (non-object) info.

To enhance wanted (object) information.

Machine & Hybrid Intelligence Lab

25 of 127

Image Filtering

25

Purpose: To suppress unwanted (non-object) info.

To enhance wanted (object) information.

Enhancive: For enhancing edges, regions.

For intensity scale standardization.

For correcting background variation.

Suppressive: Mainly for suppressing random noise.

Machine & Hybrid Intelligence Lab

26 of 127

Filters - Core component of DEEP LEARNING

26

Machine & Hybrid Intelligence Lab

27 of 127

Filters - Core component of DEEP LEARNING

27

Machine & Hybrid Intelligence Lab

28 of 127

Filters - Core component of DEEP LEARNING

28

Machine & Hybrid Intelligence Lab

29 of 127

Core DL�Filtering

29

Machine & Hybrid Intelligence Lab

30 of 127

Image Filtering

30

Machine & Hybrid Intelligence Lab

31 of 127

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0


	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Credit: S. Seitz

Image filtering

1

Machine & Hybrid Intelligence Lab

32 of 127

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0


	0	10

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

Machine & Hybrid Intelligence Lab

33 of 127

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0


0	10	20

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

Machine & Hybrid Intelligence Lab

34 of 127

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0


0	10	20	30

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

Machine & Hybrid Intelligence Lab

35 of 127


0	10	20	30	30

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

Machine & Hybrid Intelligence Lab

36 of 127


0	10	20	30	30

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

?

Machine & Hybrid Intelligence Lab

37 of 127


0	10	20	30	30




			50

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

?

Machine & Hybrid Intelligence Lab

38 of 127

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0


0	10	20	30	30	30	20	10
0	20	40	60	60	60	40	20
0	30	60	90	90	90	60	30
0	30	50	80	80	90	60	30
0	30	50	80	80	90	60	30
0	20	30	50	50	60	40	20
10	20	30	30	30	30	20	10
10	10	10	0	0	0	0	0

Image filtering

1

Credit: S. Seitz

Machine & Hybrid Intelligence Lab

39 of 127

Median Filters (non-linear)

39

A Median Filter operates over a window by selecting the median intensity in the window.

Machine & Hybrid Intelligence Lab

40 of 127

Median Filters (non-linear)

40

Machine & Hybrid Intelligence Lab

41 of 127

Filtering X-ray

41

Radiography of the skull, b. low-pass filter with a Gaussian filter (std=15, 20 x 20), c. high-pass Filter obtained from subtracting b from a.

Machine & Hybrid Intelligence Lab

42 of 127

Unsharp Masking�Lower Noise, Higher Contrast

42

histogram

Machine & Hybrid Intelligence Lab

43 of 127

Unsharp Masking

Not only noise removal, but edge enhancement is necessary!

43

Machine & Hybrid Intelligence Lab

44 of 127

Unsharp Masking

Not only noise removal, but edge enhancement is necessary!

44

Smoothed image

(low pass)

Edge enhanced image

(high pass)

Machine & Hybrid Intelligence Lab

45 of 127

Unsharp Masking

Not only noise removal, but edge enhancement is necessary!

Reminder: Edges are located in high frequency of the images!

45

Smoothed image

(low pass)

Edge enhanced image

(high pass)

Machine & Hybrid Intelligence Lab

46 of 127

Hand X-ray Unsharp Masking (alpha=0.5)

46

Original Image

Enhanced Image

Machine & Hybrid Intelligence Lab

47 of 127

Unsharp Masking: Example CT (head, axial)

47

Original CT Data

Filtered CT Data

Machine & Hybrid Intelligence Lab

48 of 127

Adaptive Filtering: Example head MRA

48

MIP of MRA data before filtering

MIP of MRA data after filtering

Machine & Hybrid Intelligence Lab

49 of 127

Adaptive Filtering: Example brain MRI

49

Original brain MRI

Enhanced brain MRI

Note the improved contrast between brain and CSF (cerebrospinal fluid)

Machine & Hybrid Intelligence Lab

50 of 127

50

Enhanced brain MRI

Note the improved contrast between brain and CSF (cerebrospinal fluid)

Adaptive Filtering: Example brain MRI (zoom)

Machine & Hybrid Intelligence Lab

51 of 127

Smoothing MRI (Denoising)

51

Machine & Hybrid Intelligence Lab

52 of 127

52

Machine & Hybrid Intelligence Lab

53 of 127

53

Machine & Hybrid Intelligence Lab

54 of 127

How to measure for evaluating noise removal algorithms?

54

SNR: Signal to Noise Ratio

CNR: Contrast to Noise Ratio

Machine & Hybrid Intelligence Lab

55 of 127

How to measure for evaluating noise removal algorithms?

55

SNR: Signal to Noise Ratio

CNR: Contrast to Noise Ratio

Machine & Hybrid Intelligence Lab

56 of 127

How to measure for evaluating noise removal algorithms?

56

SNR: Signal to Noise Ratio

CNR: Contrast to Noise Ratio

Machine & Hybrid Intelligence Lab

57 of 127

How to measure for evaluating noise removal algorithms?

57

SNR: Signal to Noise Ratio

CNR: Contrast to Noise Ratio

CNR=6

Machine & Hybrid Intelligence Lab

58 of 127

58

SNR (S/N) (of normal brain) and CNR (C/N) of multiple sclerosis plaques to normal brain on spin-density and T1 magnetic resonance images.

Machine & Hybrid Intelligence Lab

59 of 127

Thresholding

59

Machine & Hybrid Intelligence Lab

60 of 127

Example: CT Scan

60

Machine & Hybrid Intelligence Lab

61 of 127

Example: CT Scan

61

Machine & Hybrid Intelligence Lab

62 of 127

Example: CT Scan

62

Machine & Hybrid Intelligence Lab

63 of 127

Example: CT Scan

63

Machine & Hybrid Intelligence Lab

64 of 127

Example: CT Scan

64

Machine & Hybrid Intelligence Lab

65 of 127

Thresholding Methods

Huang
Intermode
Isodata
Li
MaxEntropy
Mean
MinError
Otsu
Percentile
RenyiEntropy
Moments
…

65

Machine & Hybrid Intelligence Lab

66 of 127

Example for PET Thresholding

66

ITM for tumor segmentation/FDG PET

Machine & Hybrid Intelligence Lab

67 of 127

Example for PET Thresholding

67

Machine & Hybrid Intelligence Lab

68 of 127

Further Thresholding Example – CT Bones

68

Machine & Hybrid Intelligence Lab

69 of 127

Further Thresholding Example – CT Bones

69

Machine & Hybrid Intelligence Lab

70 of 127

Head-Neck CT – Thresholding for Skull Modeling

70

(Slice Credit: P.Seutens)

Segmentation of the skull and the mandibula in CT images using thresholding. (a) Original CT image of the head. (b) Result with a threshold value of 276 Hounsfield units. The segmented bony structures are represented in color. (c) 3D rendering of the skull shows a congenital growth deficiency of the mandibula in this 8-year-old patient. This information was used preoperatively to plan a repositioning of the mandibula.

Machine & Hybrid Intelligence Lab

71 of 127

Multiple Thresholds – MRI Thresholding

71

Thresholding can be done interactively and separates the image into different regions. Valleys in the histogram indicate potentially useful threshold values

Credit: Toeonies, K.

Machine & Hybrid Intelligence Lab

72 of 127

MRI Inhomogeneity Correction �(Bias Field Correction)

72

Machine & Hybrid Intelligence Lab

73 of 127

Magnetic Field Inhomogeneity

73

Field inhomogeneity is measured in

parts per million (ppm)

with respect to the external field

Different tissues have different magnetic susceptibilities

⇨ distortions in magnetic field

distortions are most noticeable near air-tissue interfaces

Credit: K.Friston

Machine & Hybrid Intelligence Lab

74 of 127

MR Intensity Inhomogeneity

Often hardly noticeable, but registration, segmentation, and thus quantification processes are significantly affected from inhomogeneity field

74

(credit: R.Gupta)

Machine & Hybrid Intelligence Lab

75 of 127

Intensity Inhomogeneity Correction Methods

75

Original Image

Inhomogeneity Field

Corrected Image

Machine & Hybrid Intelligence Lab

76 of 127

Bias Correction Approaches

Numerous methods have been published in the last three decades

Prospective Approaches

Phantom
Multicoil
Special sequence

Retrospective Approaches

Filtering
Surface Fitting (intensity or gradient)
Segmentation (ML, MAP, FCM, nonparametric,…)
Histogram

High frequency maximization
Information maximization
Histogram matching

76

Machine & Hybrid Intelligence Lab

77 of 127

Prospective Approaches-Filtering/Surface Fitting

Rat brain MRI: a. Original image. b: after inhomogeneity

correction with the phantom based correction algorithm.

Credit: Hui et al, JMRI 2010.

Machine & Hybrid Intelligence Lab

78 of 127

78

Original

Corrected

Machine & Hybrid Intelligence Lab

79 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation)?

79

Machine & Hybrid Intelligence Lab

80 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation):

80

Machine & Hybrid Intelligence Lab

81 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation):

For a given tissue class C, CV shows how much intensity inhomogeneity is introduced (existed)

81

Machine & Hybrid Intelligence Lab

82 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation):

For a given tissue class C, CV shows how much intensity inhomogeneity is introduced (existed)
There are some drawbacks in using CV!

82

Machine & Hybrid Intelligence Lab

83 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation):

For a given tissue class C, CV shows how much intensity inhomogeneity is introduced (existed)
There are some drawbacks in using CV!

Single tissue class is used C, (alternatively Coef. Of Joint Variation can be used)

83

Machine & Hybrid Intelligence Lab

84 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation):

For a given tissue class C, CV shows how much intensity inhomogeneity is introduced (existed)
There are some drawbacks in using CV!

Single tissue class is used C, (alternatively Coef. Of Joint Variation can be used)
Sensitive to brightness of the image (change the mean, not the std)

84

Machine & Hybrid Intelligence Lab

85 of 127

Coefficient of Variation as a quantitative evaluation metric

CV (coefficient of variation):

For a given tissue class C, CV shows how much intensity inhomogeneity is introduced (existed)
There are some drawbacks in using CV!

Single tissue class is used C, (alternatively Coef. Of Joint Variation can be used)
Sensitive to brightness of the image (change the mean, not the std)
…

85

Machine & Hybrid Intelligence Lab

86 of 127

MRI Intensity Standardization

86

Machine & Hybrid Intelligence Lab

87 of 127

MR Intensity Non-Standardness

Acquisition-to-acquisition signal intensity variations (non-standardness) are inherent in MR images.

87

Machine & Hybrid Intelligence Lab

88 of 127

MR Intensity Non-Standardness

Acquisition-to-acquisition signal intensity variations (non-standardness) are inherent in MR images.

88

Machine & Hybrid Intelligence Lab

89 of 127

MR Intensity Non-Standardness

What is changed?

PD PD T2 T2

89

Machine & Hybrid Intelligence Lab

90 of 127

MR Intensity Non-Standardness

Intensity inhomogeneity is removed! (N3 or N4 can be used).

PD PD T2 T2

90

Machine & Hybrid Intelligence Lab

91 of 127

What is changed now?

91

Machine & Hybrid Intelligence Lab

92 of 127

Intensities are standardized!

92

Machine & Hybrid Intelligence Lab

93 of 127

Intensity Non-Standardness

MR image intensities do not possess a tissue-specific numeric meaning even in images acquired for

the same subject,
on the same scanner,
for the same body region,

by using the same pulse sequence

93

Machine & Hybrid Intelligence Lab

94 of 127

Intensity Non-Standardness

94

same brain slice,

same person,

same scanners,

different imaging times,

intensities are significantly

different for the same

Tissue type!

MR image intensities do not possess a tissue-specific numeric meaning even in images acquired for

the same subject,
on the same scanner,
for the same body region,

by using the same pulse sequence

Machine & Hybrid Intelligence Lab

95 of 127

95

10 Different

PD studies

ORIGINAL

HISTOGRAMS

Credit: L. Nyul

Machine & Hybrid Intelligence Lab

96 of 127

96

10 Different

PD studies

ORIGINAL

HISTOGRAMS

BACKGROUND

FOREGROUND

Credit: L. Nyul

Machine & Hybrid Intelligence Lab

97 of 127

97

10 Different

PD studies

ORIGINAL

HISTOGRAMS

BACKGROUND

FOREGROUND

AFTER STANDARDIZATION

Credit: L. Nyul

Machine & Hybrid Intelligence Lab

98 of 127

98

Original

Gray Scale

After Intensity

Standardization

Credit: L. Nyul

Machine & Hybrid Intelligence Lab

99 of 127

How to Standardize MRI Intensities?

99

Machine & Hybrid Intelligence Lab

100 of 127

How to Standardize MRI Intensities?

Histogram

100

Machine & Hybrid Intelligence Lab

101 of 127

How to Standardize MRI Intensities?

Histogram

101

Mono-modal

Bi-modal

Machine & Hybrid Intelligence Lab

102 of 127

How to Standardize MRI Intensities?

Histogram

102

Mono-modal

Bi-modal

Minimum

intensity

Maximum

intensity

Second mode

Of the histogram

Shoulder of the

Background hump

Minimum and maximum

Percentile intensities

Machine & Hybrid Intelligence Lab

103 of 127

How to Standardize MRI Intensities?

Histogram

103

Mono-modal

Bi-modal

Minimum

intensity

Maximum

intensity

Second mode

Of the histogram

Shoulder of the

Background hump

Minimum and maximum

Percentile intensities

Machine & Hybrid Intelligence Lab

104 of 127

How to Standardize MRI Intensities?

Histogram

104

Mono-modal

Bi-modal

Minimum

intensity

Maximum

intensity

Second mode

Of the histogram

Shoulder of the

Background hump

Minimum and maximum

Percentile intensities

Machine & Hybrid Intelligence Lab

105 of 127

How to Standardize MRI Intensities?

Histogram

105

Mono-modal

Bi-modal

Minimum

intensity

Maximum

intensity

Second mode

Of the histogram

Shoulder of the

Background hump

Minimum and maximum

Percentile intensities

Machine & Hybrid Intelligence Lab

106 of 127

How to Standardize MRI Intensities?

Histogram

106

Mono-modal

Bi-modal

Minimum

intensity

Maximum

intensity

Second mode

Of the histogram

Shoulder of the

Background hump

Minimum and maximum

Percentile intensities

Machine & Hybrid Intelligence Lab

107 of 127

Map background and foreground into the fixed intensity regions!

107

🡨Location of different modes

Machine & Hybrid Intelligence Lab

108 of 127

Map background and foreground into the fixed intensity regions!

108

Fixed (standardized modes)

Machine & Hybrid Intelligence Lab

109 of 127

109

IMAGE SCALE

STANDARD SCALE

m1i

p1i

μ_i

p2i

m2i

s’_1i

s’_2i

s1

s2

μ_s

Machine & Hybrid Intelligence Lab

110 of 127

110

IMAGE SCALE

STANDARD SCALE

m1i

p1i

μ_i

p2i

m2i

s’_1i

s’_2i

s1

s2

μ_s

(known, fixed)

Machine & Hybrid Intelligence Lab

111 of 127

111

IMAGE SCALE

STANDARD SCALE

m1i

p1i

μ_i

p2i

m2i

s’_1i

s’_2i

s1

s2

μ_s

(known, fixed)

for a given image i,

we must calculate

these parameters from

histogram and then

transform them into

the standard scale

values.

Machine & Hybrid Intelligence Lab

112 of 127

112

IMAGE SCALE

STANDARD SCALE

m1i

p1i

μ_i

p2i

m2i

s’_1i

s’_2i

s1

s2

μ_s

(known, fixed)

Machine & Hybrid Intelligence Lab

113 of 127

113

IMAGE SCALE

STANDARD SCALE

m1i

P1i x

μ_i

p2i

m2i

s’_1i

s’_2i

s1

s2

μ_s

(known, fixed)

Machine & Hybrid Intelligence Lab

114 of 127

114

IMAGE SCALE

STANDARD SCALE

m1i

P1i x

μ_i

p2i

m2i

s’_1i

s’_2i

s1

s2

μ_s

(known, fixed)

Machine & Hybrid Intelligence Lab

115 of 127

Intensity Mapping Function

115

Machine & Hybrid Intelligence Lab

116 of 127

Standardized Intensity Mapping T1-MRI

116

Original Scale

Standard Scale

Machine & Hybrid Intelligence Lab

117 of 127

How to determine standardized parameters?

117

One way is to have some training images,

and find average values (or median) for

Histogram parameters

Machine & Hybrid Intelligence Lab

118 of 127

Foot MRI Intensity Standardization

118

Original Scale

Standard Scale

Machine & Hybrid Intelligence Lab

119 of 127

Different Body region, MR Modality, and Subjects for Training? (mixed training)

119

Machine & Hybrid Intelligence Lab

120 of 127

Quantitative Comparisons

120

Credit: L. Nyul

Machine & Hybrid Intelligence Lab

121 of 127

Interplay between Denoising, Bias Correction, and Intensity Standardization ?

121

Denoising

Intensity Standardization

Bias Correction

Intensity Standardization

Bias Correction

Denoising

Bias Correction

Denoising

Intensity Standardization

Bias Correction

Intensity Standardization

Denoising

(A)

(B)

(C)

(D)

Machine & Hybrid Intelligence Lab

122 of 127

Interplay between Denoising, Bias Correction, and Intensity Standardization ?

122

Denoising

Intensity Standardization

Bias Correction

Intensity Standardization

Bias Correction

Denoising

Bias Correction

Denoising

Intensity Standardization

Bias Correction

Intensity Standardization

Denoising

(A)

(B)

(C)

(D)

Machine & Hybrid Intelligence Lab

123 of 127

Effects of Intensity Standardization on Image Registration

The results from literature (Bagci PRL 2010) imply that the accuracy of image registration not only depends on spatial and geometric similarity but also on the similarity of the intensity values for the same tissues in different images

123

Machine & Hybrid Intelligence Lab

124 of 127

Effects of Intensity Standardization on Image Segmentation

Y.Zhuge et al (CVIU 2009) showed that intensity standardization simplifies brain image segmentation

124

Before standardization

After standardization

Machine & Hybrid Intelligence Lab

125 of 127

Summary

ROI, VOI, SNR, CNR, Resolution, Noise, Thresholding, Histogram…are all important terminologies to know
Intensity non-standardness is inherent in MRI

must standardize

Intensity standardization + inhomogeneity correction + denoising need to be handled prior to image analysis
Intensity non-standardness affects

Perception/qualitative analysis
Image analysis/quantification

Segmentation
Registration
Recognition/localization

125

Machine & Hybrid Intelligence Lab

126 of 127

References and Slide Credits

P. Suetens, Fundamentals of Medical Imaging, Cambridge Univ. Press.
ITK.org
siemens.com
slicer.org
Must watch lecture (basic): https://www.youtube.com/watch?v=jWRIKNeCXjI
http://www.medicine.uodiyala.edu.iq/uploads/lectures/physics/MRI_Physics.pdf
MRI Made Easy (free, easy book): http://www.stat.columbia.edu/~martin/Tools/MRI_Made_Easy.pdf
www.paperswithcode.com
Wikipedia
Radiologycafe
Dennis Shanon Hill slideshare
Slides from Jayaram K. Udupa, MIPG of University of Pennsylvania, PA.
N. Bryan, Intro. to the science of medical imaging, Cambridge Univ. Press.
DeepSeek and GPT for grammar and flow correction.

126

Machine & Hybrid Intelligence Lab

127 of 127

127

Thank you for listening!

Ulas Bagci, Ph.D.,

Associate Professor,

Director of Machine & Hybrid Intelligence Lab,

Department of Radiology, Feinberg School of Medicine,

Department of Biomedical Engineering (Courtesy),

Department of Electrical and Computer Engineering (Courtesy)

Northwestern University,

737 N. Michigan Avenue Suite 1600,

Chicago, IL 60611, USA

Phone: +1 312-694-4951�Cell: +1 240 383 8587�

Web: https://www.bagcilab.com

Machine & Hybrid Intelligence Lab

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0

0	0	0	0	0	0
0	0	0	0	0	0
0	90	90	90	90	90
0	90	90	90	90	90
0	90	90	90	90	90
0	90	0	90	90	90
0	90	90	90	90	90
0	0	0	0	0	0
90	0	0	0	0	0
0	0	0	0	0	0