UNIT IV

REGISTRATION AND VISUALISATION

QUESTION BANK
2 MARKS
1. Define Image Registration.

Image Registration is a crucial image processing technique that involves aligning multiple
images of the same scene taken at different times, from different viewpoints, or using different
sensors. The primary goal is to establish spatial correspondence between these images,
ensuring that corresponding features or objects align precisely in a common coordinate
system.

2. What are the two types of image registration?

The two primary types of image registration are:

 Intensity-Based Methods: These methods directly compare the pixel intensities of the
images to find the best alignment. They are suitable when images have subtle
differences or lack distinct features.
Common techniques include:
 Correlation: Measures the similarity between image regions.
 Mutual Information: Evaluates the statistical dependence between image
intensities.
 Feature-Based Methods: These methods identify distinctive features (e.g., corners,
edges) in the images and establish correspondences between them. They are robust to
noise and variations in intensity.
Common steps include:
 Feature Detection: Extracting key points from the images.
 Feature Description: Creating descriptors for each feature.
 Feature Matching: Finding corresponding features between images.
 Transformation Estimation: Determining the transformation that aligns the
images based on the feature matches.

3. What is rigid body in image processing?

In image processing, a rigid body is a model that assumes that the distances and internal
angles of an object in an image do not change during registration. This model is used to
describe objects that behave like rigid bodies in the real world, such as individual bones
and the brain.A rigid body is a system of particles where the distance between the particles
does not change, even when a force is applied.
A rigid body can move in two ways:
 Translational motion: Every point in the body moves the same distance parallel to
a line.
 Rotational motion: One point in the body is fixed, and the body rotates.
4. Discuss the challenges of registering images from different modalities, such as MRI
and CT.

Challenges of Registering Images from Different Modalities (e.g., MRI and CT):
  Intensity Inhomogeneity: Different modalities capture different physical properties of
tissues, leading to significant variations in image intensity. This makes it difficult to
directly compare pixel values between images.
 Spatial Mismatch: Modalities often have different spatial resolutions and imaging
planes, resulting in anatomical structures appearing at different locations or with varying
degrees of detail.
 Partial Volume Effects: Partial volume averaging can cause blurring and loss of detail at
tissue boundaries, further complicating the registration process.
 Noise and Artifacts: Noise and artifacts present in one or both modalities can interfere
with feature extraction and matching.

5. Propose few methods to overcome the challenges caused in registering images from
different modalities.

a. Intensity Normalization:
 Histogram Matching: Adjust the intensity distributions of the images to make them
more comparable.
 Non-rigid Registration with Intensity Correction: Incorporate intensity correction
terms into the registration algorithm to account for modality-specific intensity
variations.
b. Feature-Based Registration with Robust Matching:
 Mutual Information: A popular metric that measures the statistical dependence
between images, less sensitive to intensity variations.
 Robust Feature Descriptors: Utilize feature descriptors that are more invariant to
intensity changes and noise, such as SIFT or SURF.
 Robust Matching Strategies: Employ techniques like RANSAC (Random Sample
Consensus) to identify and discard outlier matches caused by noise or artifacts.

6. Discuss the role of image registration in image-guided surgery.

The key roles of Image Registration in Image-Guided Surgery are as follows:

a. Preoperative Planning:
 Treatment Planning: By registering preoperative images (e.g., CT, MRI) with surgical
plans, surgeons can accurately plan the surgical approach, determine the optimal
incision site, and anticipate potential challenges.
 Simulation: Image registration allows for the simulation of surgical procedures on the
preoperative images, helping surgeons to refine their techniques and anticipate potential
complications.

b. Intraoperative Guidance:
 Navigation: By registering preoperative images with real-time intraoperative images
(e.g., fluoroscopy, ultrasound), surgeons can navigate instruments and devices to the
target location with high precision.
 Visualization: Image registration enables the overlay of preoperative images onto real-
time images, providing surgeons with a comprehensive view of the patient's anatomy
and the location of the target.

7. Discuss the role of image registration in radiation therapy.

The key roles of Image Registration in Radiation Therapy are as follows:

a. Treatment Planning:
  Target Localization: By registering CT and MRI images with the patient's anatomy,
radiation oncologists can accurately delineate the tumor and surrounding critical
structures.
 Dose Calculation: Image registration ensures that the radiation dose is accurately
calculated and delivered to the target volume while sparing healthy tissues.

b. Treatment Delivery:
 Image-Guided Radiation Therapy (IGRT): By registering real-time images (e.g.,
cone-beam CT) with the treatment plan, radiation oncologists can monitor the
patient's position and adjust the treatment delivery in real-time to compensate for
any movement or changes in anatomy.

8. Analyze the impact of registration errors on the accuracy of medical image analysis.

Registration errors in medical image analysis can have significant consequences, impacting
the accuracy and reliability of diagnostic and therapeutic procedures. Here's an analysis of the
potential impacts:

a. Misdiagnosis and Treatment Errors:

 Inaccurate Tumor Localization: If registration errors lead to incorrect localization of
tumors or other lesions, it can result in misdiagnosis, incorrect treatment planning,
and potentially ineffective or even harmful therapies.
 Incorrect Radiation Doses: Inaccurate registration in radiation therapy can lead to
the misdelivery of radiation doses, potentially exposing healthy tissues to excessive
radiation or failing to adequately target the tumor.

b. Reduced Confidence in Analysis Results:

 Questionable Segmentation: Registration errors can lead to inaccurate segmentation
of anatomical structures, impacting the reliability of quantitative measurements and
the accuracy of computer-aided diagnosis systems.
 Erroneous Motion Correction: In functional imaging (e.g., fMRI), inaccurate motion
correction can lead to artifacts and distortions in the data, making it difficult to
interpret brain activity patterns.

c. Increased Costs and Resource Utilization:

 Repeat Procedures: Registration errors may necessitate repeat procedures, such
as additional imaging scans or surgical interventions, leading to increased costs and
patient discomfort.
 Wasted Resources: Incorrectly registered images can lead to wasted time and
resources in the analysis and interpretation process.

9. Explain the advantages and disadvantages of perspective projections in medical

imaging.

The advantages and disadvantages of Perspective Projection are:

Advantages:

 Provides depth cues: Creates a more realistic and visually appealing image, with objects
appearing smaller as they recede into the distance.
 Facilitates depth perception: Helps viewers to better understand the 3D spatial
relationships between objects.
Disadvantages:

 Distorts object size and shape: Objects appear smaller and distorted as they move
further away from the viewer.
 Measurements can be inaccurate: Direct measurements from the image may not
accurately reflect the true dimensions of the object.

10. Discuss the challenges of visualizing 3D medical images.

Challenges of Visualizing 3D Medical Images are as follows:

a. Data Volume and Complexity: Medical image datasets can be massive, containing
terabytes of data. 3D rendering and manipulation of such large datasets can be
computationally expensive and time-consuming.Complex anatomical structures with
intricate details pose significant challenges for accurate visualization and interpretation.
b. Perception and Cognitive Load: Human perception is limited in interpreting 3D information
from 2D displays. Navigating and understanding complex 3D structures can be cognitively
demanding. Overlapping structures and cluttered visual scenes can hinder accurate
interpretation.
c. Data Variety and Heterogeneity: Medical images come from various modalities (CT, MRI,
PET, etc.), each with unique characteristics and challenges for visualization. Integrating and
visualizing data from multiple modalities can be complex.

11. Compare and contrast the performance of surface-based rendering and volume
rendering for visualizing different types of medical images.

SURFACE BASED RENDERING VOLUME BASED RENDERING

Identifies and extracts surfaces of interest (e.g., organ Renders the entire volume data directly, assigning
boundaries) from the volume data. These surfaces are colours and opacities to individual voxels based on
then represented as 3D models (polygons) and their intensity values.Allows for visualization of internal
rendered using traditional computer graphics structures and varying tissue densities.
techniques.
Advantages: Advantages:

 Fast Rendering: Generally faster than volume  Visualizes Internal Structures: Provides
rendering, especially for complex scenes. insights into the internal anatomy and tissue
 Good for Well-Defined Surfaces: Excellent for properties.
visualizing objects with clear boundaries, such  Flexible: Allows for various rendering
as organs with distinct edges. techniques (e.g., maximum intensity projection,
 Interactive Exploration: Enables smooth and minimum intensity projection, shaded surface
interactive manipulation and rotation of 3D display) to highlight different features.
models.  No Segmentation Required: Eliminates the
need for explicit segmentation of structures.

Disadvantages: Disadvantages:

 Limited Internal Structures: Only displays the  Slower Rendering: Can be computationally
surfaces of objects, providing no information expensive, especially for large datasets.
about internal structures.  Requires More Powerful Hardware: Demands
 Segmentation Challenges: Requires accurate significant computational resources for real-
segmentation of the desired structures, which time rendering.
can be challenging and time-consuming.
12. Analyze the impact of visualization techniques on the accuracy of medical diagnosis
and treatment planning.

Medical image visualization techniques significantly impact diagnostic accuracy and treatment
planning. By providing 3D representations of internal structures, these methods enhance lesion
detection, disease characterization, and surgical planning. For instance, volume rendering
allows for the visualization of internal structures and tissue densities, aiding in the identification
of subtle abnormalities. Surface rendering provides clear depictions of organ boundaries,
facilitating precise surgical planning and minimizing invasiveness. These techniques also
improve communication and collaboration among healthcare professionals, leading to more
informed treatment decisions and better patient outcomes. However, it's crucial to use
visualization techniques judiciously, ensuring proper training and interpretation to minimize
potential misinterpretations and maximize their benefits.

13. Discuss the role of visualization in medical education and patient communication.

Visualization plays a vital role in both medical education and patient communication. In medical
education, 3D models, simulations, and interactive visualizations enhance student
understanding of complex anatomical structures and physiological processes. This facilitates
learning and improves knowledge retention compared to traditional text-based methods.In
patient communication, visualizations help bridge the gap between medical jargon and patient
comprehension. 3D models of organs, disease progression, and surgical procedures can
empower patients to understand their conditions and treatment options better. This improved
understanding fosters better patient-physician communication, enhances patient engagement,
and improves treatment adherence.

14. Implement an orthogonal projection algorithm in Matlab using the imrotate function.

function projected_image = orthogonal_projection(image, angle)

#ORTHOGONAL_PROJECTION Performs orthogonal projection of an image.

rotated_image = imrotate(image, angle, 'nearest');

#Crop the rotated image to remove any empty borders

[rows, cols] = size(rotated_image);

center_row = round(rows / 2);

center_col = round(cols / 2);

min_row = max(1, center_row - floor(size(image, 1) / 2));

max_row = min(rows, center_row + ceil(size(image, 1) / 2));

min_col = max(1, center_col - floor(size(image, 2) / 2));

max_col = min(cols, center_col + ceil(size(image, 2) / 2));

projected_image = rotated_image(min_row:max_row, min_col:max_col);

end
15. Implement a perspective projection algorithm in Matlab using the imwarp function.

# PERSPECTIVE_PROJECTION Performs perspective projection of an image.

# Create a projective transformation object

tform = fitgeotrans(source_points, destination_points, 'projective');

# Perform perspective projection

projected_image = imwarp(image, tform, 'OutputView', imref2d(size(destination_points)));

end

16. What are the four steps every method of image registration has to go through for
image alignment?

There are major four steps that every method of image registration has to go through for image
alignment. These could be listed as follows:

 Feature detection: A domain expert detects salient and distinctive objects (closed
boundary areas, edges, contours, line intersections, corners, etc.) in both the reference
and sensed images.
 Feature matching: It establishes the correlation between the features in the reference
and sensed images. The matching approach is based on the content of the picture or
the symbolic description of the control point-set.
 Estimating the transform model: The parameters and kind of the so-called mapping
functions are calculated, which align the detected picture with the reference image.
 Image resampling and transformation: The detected image is changed using mapping
functions.

17. Define orthographic projection.

Orthographic projection is a form of parallel projection in which the top, front, and side of an
object are projected onto perpendicular planes. All 3 views are shown in the final orthogonal
sketch. An isometric projection is one 3D image drawn on an isometric grid.

18. Explain the advantages and disadvantages of orthogonal projections in medical

imaging.

The advantages and disadvantages of Orthogonal Projections are:

Advantages:

 Preserves true dimensions: Objects appear in their actual size and shape, without
distortion.
 Simple to interpret: Easy to visualize and understand spatial relationships.
 Ideal for measurements: Accurate measurements of length, width, and angles can be
made directly from the image.

Disadvantages:

 Limited depth perception: Can be challenging to perceive the depth of objects or their
relative positions in 3D space.
  May require multiple views: Often requires multiple projections from different angles to
fully understand the object's 3D structure.

19. Represent the block diagram of volume rendering.

20. Find the type of algorithm used in the below figure(a) and explain.

figure (a)

Ray Sum technique is used. It is a technique whereby hypothetical X-ray are sent from each
pixel of a source towards a volume to a final image in the screen. The objective is to sum all
the ray lengths through all the voxels in CT volume data multiplied by voxel densities to get the
radiological path.

The values associated with the voxels determine what happens to each ray and therefore what
image is finally reconstructed. For each pixel of the final image on the screen, a ray is used to
intersect parallel voxels in the volume. Each pixel of the projection gets a 12-bit value by
averaging all intensities of the intersected pixels.
16 MARKS
1. Highlight the significance of registration of various imaging modalities and explain
the concepts of image visualization in healthcare using Matlab.

Significance of Registration of Imaging Modalities:

Image registration is a critical process in medical imaging that involves aligning images from
different sources or modalities to achieve a common spatial reference. Its significance in
healthcare includes:

a. Enhanced Diagnosis:
 Combining information from multiple modalities, such as CT (anatomical detail) and
PET (functional detail), improves diagnostic accuracy.
 Registered images provide a comprehensive view of diseases, such as tumors, by
overlaying functional and structural details.

b. Treatment Planning:
 In radiotherapy, accurate registration ensures precise targeting of tumors while
sparing healthy tissues.
 Fusion of MRI and CT helps in pre-surgical planning by accurately delineating
anatomical structures.

c. Monitoring and Follow-Up:

 Registration of images taken at different times enables tracking of disease
progression or response to treatment.

d. Multimodal Data Integration:

 Different imaging techniques provide complementary information. Registration aligns
them to create a unified dataset.

Concepts of Image Visualization in Healthcare Using MATLAB:

MATLAB is a powerful tool for visualizing medical images, enabling healthcare professionals
and researchers to analyze, process, and interpret imaging data effectively.

i. Reading and Displaying Images:

Use MATLAB's imread and imshow functions to load and display medical images (e.g., X-rays,
MRI, CT scans).

matlab
img = imread('CT_Scan.jpg');
imshow(img, []);
title('CT Scan Image');

ii. Image Enhancement:

Techniques like contrast adjustment (imadjust), histogram equalization (histeq), and noise
reduction improve the visibility of features in medical images.
iii. 3D Volume Visualization:

For 3D imaging modalities (e.g., MRI, CT), MATLAB's volshow or slice functions enable
visualization of the volumetric data.

matlab
volumeData = load('MRI_Volume.mat');
volshow(volumeData);

iv. Multimodal Image Fusion:

Overlay images from different modalities for better analysis. For instance, combining a PET
scan over a CT scan:

matlab
fusedImage = imfuse(CT_image, PET_image, 'blend');
imshow(fusedImage);

v. ROI (Region of Interest) Analysis:

Define and analyze specific areas in an image using tools like roipoly for focused evaluation.

vi. Segmentation and Feature Extraction:

Segment specific structures (e.g., tumors) using methods like activecontour or edge.

matlab
bw = edge(img, 'Canny');
imshow(bw);
title('Edge Detection for Feature Extraction');

vii. Visualization of Registration Results:

Display registered images side by side or superimposed to validate the alignment.

matlab
registeredImage = imregister(movingImage, fixedImage, 'rigid', optimizer, metric);
imshowpair(fixedImage, registeredImage);

2. Explain the significance of registration of various imaging modalities in healthcare.

Discuss how the registration of images from different modalities such as MRI, CT,
and PET can facilitate multi-modal image analysis and improve diagnostic accuracy.

Significance of Registration of Various Imaging Modalities in Healthcare

Image registration refers to the process of aligning two or more images of the same scene
taken at different times, from different perspectives, or using different imaging modalities. In
healthcare, it plays a critical role in diagnostics, treatment planning, and research by ensuring
that images from various modalities are spatially aligned to a common reference.
Importance of Image Registration in Healthcare are:

a. Comprehensive Diagnosis:

Different imaging modalities capture complementary information:

 MRI provides detailed soft-tissue contrast.

 CT offers high-resolution anatomical details.
 PET reveals metabolic and functional activity.

Image registration combines these modalities, enabling clinicians to analyze structural and
functional data simultaneously, leading to better diagnosis of conditions like tumors, brain
disorders, and cardiac issues.

b. Improved Treatment Planning:

Accurate alignment of multimodal images is essential in planning radiation therapy, as it

ensures precise targeting of cancerous tissues while avoiding damage to adjacent healthy
structures. For example, combining CT for anatomy and PET for tumor metabolism allows
oncologists to fine-tune treatment plans.

c. Monitoring Disease Progression:

Registration enables comparison of images taken at different times or from different

modalities, helping clinicians monitor disease progression or evaluate the effectiveness of
treatment.

d. Guidance During Surgeries:

Registered images help surgeons navigate complex anatomical regions. For instance,
overlaying MRI on CT can provide both detailed anatomy and pathological insights during
brain surgeries.

e. Integration of Multimodal Data:

Registration integrates diverse data from modalities, allowing advanced image analysis, such
as machine learning-based diagnostic models that utilize combined information from MRI, CT,
and PET.

f. Support for Research and Innovation:

Image registration is a cornerstone for developing cutting-edge techniques like 3D printing of

organs, augmented reality in surgery, and AI-based diagnostic tools.
Registration of MRI, CT, and PET for Multi-Modal Image Analysis are:

a. MRI and CT Registration:

MRI excels in soft tissue contrast but lacks detail in bone structures. CT provides
excellent resolution for bones and hard tissues. Registration of MRI and CT is widely used
in:Orthopedics: Fusion of bone (CT) and ligament or cartilage (MRI) images. Neurosurgery:
Combining soft tissue (MRI) and skull (CT) information.

b. PET and CT Registration:

PET highlights functional or metabolic activity but lacks anatomical detail. CT provides the
anatomical context.

Registration facilitates:

 Cancer Diagnosis: PET identifies metabolically active cancerous tissues, and CT

localizes them anatomically.
 Cardiology: PET-CT helps in assessing both the functional and structural state of the
heart.

c. MRI and PET Registration:

MRI offers superior soft-tissue resolution, while PET provides functional data.

Applications include:

 Neuroimaging: Combining PET’s metabolic activity mapping with MRI’s structural detail
is crucial for diagnosing conditions like Alzheimer’s, epilepsy, and brain tumors.
 Oncology: Enhanced detection and staging of tumors.

Benefits of Multi-Modal Image Analysis

a. Enhanced Diagnostic Accuracy:

By integrating functional and anatomical data, clinicians can identify abnormalities with greater
precision. Example: A tumor’s exact size, location, and metabolic activity can be visualized
simultaneously.

b. Minimized Misdiagnosis:

Registration ensures that structures or abnormalities visible in one modality align correctly with
features in another, reducing diagnostic errors.

c. Patient-Specific Insights:

Multi-modal analysis tailored to individual patients helps create personalized treatment plans.

d. Improved Visualization for Decision-Making:

Fused images make it easier for clinicians to identify correlations between structural and
functional changes, leading to informed decisions.
Challenges in Image Registration

a. Complexity of Algorithms:

Different modalities produce images with varying resolutions and intensity distributions,
making alignment computationally intensive.

b. Artifacts and Noise:

Imaging artifacts and noise can affect registration accuracy.

c. Time and Resource-Intensive:

High computational resources and time are often required for precise multimodal registration.

3. Discuss the impact of noise and image artifacts on the accuracy of rigid body
registration algorithms.

Impact of Noise and Image Artifacts on the Accuracy of Rigid Body Registration Algorithms

Rigid body registration is a process where images are aligned by applying transformations
such as translation and rotation, without altering their shape or scale. The accuracy of these
algorithms depends significantly on the quality of the images involved. Noise and artifacts can
severely affect their performance, as detailed below:

a. Noise in Medical Images

Noise refers to random variations in pixel intensity that do not represent actual information. It
can arise due to limitations in imaging hardware, environmental conditions, or transmission
errors.

Types of Noise in Medical Imaging:

 Gaussian Noise: Common in MRI and CT, caused by thermal effects in sensors.

 Poisson Noise: Arises in low-light imaging or photon-counting modalities (e.g., PET).

 Speckle Noise: Present in ultrasound due to interference from sound waves.

Impact on Rigid Body Registration:

 Reduced Feature Detection: Noise obscures anatomical landmarks, making it harder to

identify points for alignment.

 Inaccurate Similarity Metrics: Algorithms often rely on metrics like Mean Squared Error
(MSE), mutual information, or cross-correlation. Noise introduces inconsistencies in
these metrics, leading to incorrect alignments.

 Convergence Issues: Optimization algorithms used in registration may converge to

local minima due to the distortions caused by noise.

 Increased Computation Time: Noise requires additional preprocessing (e.g., filtering) or

multiple iterations to achieve acceptable accuracy.
b. Image Artifacts

Artifacts are systematic distortions or anomalies in images that do not represent the true
structures being imaged. They can result from equipment malfunctions, patient motion, or
reconstruction algorithms.

Common Artifacts:

 Motion Artifacts: Caused by patient movement during image acquisition (e.g., blurring
or ghosting).
 Metallic Artifacts: Occur in CT images due to the presence of implants or prosthetics,
leading to streaks or distortions.
 Beam-Hardening Artifacts: Appear in CT scans when high-energy X-rays penetrate
dense materials, creating non-uniform intensity distributions.
 Susceptibility Artifacts: Common in MRI, caused by variations in magnetic susceptibility
near air-tissue or metal interfaces.

Impact on Rigid Body Registration:

 Mismatched Structures: Artifacts create false features or suppress true anatomical

details, leading to misalignment between images.
 Degraded Similarity Metrics: Artifacts distort intensity values, which can mislead metrics
used to evaluate the alignment.
 Loss of Local Accuracy: While global alignment may succeed, localized regions
affected by artifacts may remain misaligned.
 Patient-Specific Variability: Artifacts are often unique to individual scans, complicating
the generalization of registration algorithms.

c. Combined Effect of Noise and Artifacts

When noise and artifacts co-exist, their combined impact can exacerbate registration errors:

 Ambiguity in Feature Matching: Both distort true features, making the identification of
corresponding points in images unreliable.
 Increased Preprocessing Requirements: Advanced techniques such as denoising filters
(e.g., Gaussian filters, wavelet transforms) and artifact correction (e.g., metal artifact
reduction) are required, adding computational complexity.

d. Mitigation Strategies

To counter the negative effects of noise and artifacts, the following approaches can be
employed:

 Preprocessing: Noise reduction through filters (e.g., median, Gaussian, or anisotropic

diffusion filters). Artifact correction algorithms tailored to specific modalities (e.g.,
motion correction in MRI, metal artifact reduction in CT).
 Robust Registration Metrics: Metrics like mutual information and normalized cross-
correlation are less sensitive to noise and artifacts compared to simpler metrics like
MSE.
 Outlier Detection and Removal: Algorithms like RANSAC (Random Sample Consensus)
can identify and exclude points affected by artifacts or noise.
 Multi-Resolution Techniques: Performing registration on a coarse-to-fine hierarchy can
reduce the influence of noise and artifacts at higher resolutions.
  Regularization in Optimization: Incorporating regularization terms into the cost function
to prevent overfitting to noisy or artifact-prone regions.
 Advanced Algorithms: Incorporating machine learning techniques that are trained to
account for noise and artifacts. Using hybrid methods that combine rigid body
registration with deformable models for greater flexibility in alignment.

4. How are principal axes computed from image data, and how are they used for
registration?

Principal axes can be computed from image data using a variety of methods, often in the
context of principal component analysis (PCA) or through direct computation of the image
moments. In the context of image registration, they can help align or match images by finding
the primary orientation of an object or structure in one image and aligning it with the same
object or structure in another image.

a. Image Pre-processing:
 Grayscale Conversion: If the image is in color, it may first be converted to grayscale,
as the principal axis calculation typically deals with intensity values.
 Thresholding or Segmentation: Often, thresholding or other segmentation
techniques (like edge detection or region growing) are used to isolate the object of
interest, simplifying the computation of the principal axes by removing irrelevant
background.

b. Computing the Image Moments

 Central Moments: The first step in finding the principal axes is to compute the image
moments, specifically the central moments, which are weighted averages of pixel
intensities. For a 2D image, the central moments can be calculated as:

where I(x,y) is the intensity at pixel (x,y), and xˉ, yˉ are the centroids of the image
(the first moments).

 Covariance Matrix: The next step is to compute the covariance matrix of the pixel
coordinates weighted by their intensities (or the binary mask if segmentation is
applied). This matrix is typically of the form:

c. Eigenvalue Decomposition
 The covariance matrix is then diagonalized by finding its eigenvalues and
eigenvectors. The eigenvectors represent the principal axes, and the eigenvalues
represent the variance along each axis.
 The larger eigenvalue corresponds to the axis with the greatest spread or variance
in the data, which is typically the "major principal axis," and the smaller eigenvalue
corresponds to the "minor principal axis."
 The eigenvectors define the orientation of these axes.

d. Alignment for Registration

 Rotation and Scaling: Once the principal axes are computed for both images, they
can be used to align the images during registration. If one image’s principal axes are
 rotated or scaled compared to the other, this transformation is applied to align the
two images.
 Rigid Transformation: The principal axes are often used in a rigid transformation,
where the rotation (and optionally scaling) is applied to match the axes of one image
to the axes of another. This helps in overcoming variations in orientation, shape, or
size between the images.
 Affine or Non-Rigid Registration: In more advanced cases, the principal axes can be
used as an initial guess for affine or non-rigid registration algorithms, which refine
the alignment by optimizing over more complex transformations (like warping).

e. Applications in Image Registration

 Medical Imaging: In medical image registration (e.g., CT or MRI scans), principal
axes help align organs or anatomical structures between different image modalities
or time points.
 Computer Vision: In object recognition or 3D reconstruction, computing the principal
axes can help in aligning 3D models with 2D images or matching different views of
the same object.
 Multimodal Registration: When registering images from different modalities (e.g., CT
and MRI), principal axes can provide an initial alignment, making subsequent
registration more efficient.

Advantages of Using Principal Axes in Registration

 Robustness: Principal axes provide a statistically meaningful representation of object

orientation, making them less prone to noise and artifacts.
 Automation: The method is automated and does not require manual identification of
landmarks.
 Computational Efficiency: PCA and eigen decomposition are computationally efficient
for large datasets.
 Alignment Consistency: Principal axes ensure consistent alignment regardless of initial
object orientation.

Limitations and Challenges

 Symmetrical Shapes: For symmetrical objects, the principal axes might not uniquely
define orientation, leading to ambiguity.
 Noise Sensitivity: Heavily noisy images or those with artifacts can distort the
computation of principal axes.
 Multi-Object Images: In images with multiple overlapping objects, principal axes might
not accurately represent individual object orientations.

5. Propose a method for improving the visualization of small structures using

orthogonal projections.

Improving the Visualization of Small Structures Using Orthogonal Projection

Orthogonal projections are widely used in medical imaging to visualize 3D structures in a 2D

plane. However, small structures, such as microvessels, lesions, or fine anatomical features,
are often difficult to discern due to overlapping larger structures, low contrast, or insufficient
spatial resolution. Below is a proposed method to enhance the visualization of such small
structures using orthogonal projections:
a. Preprocessing the Data:

 Image Acquisition: Use high-resolution imaging techniques to capture fine details

(e.g., CT, MRI, micro-CT, or high-magnification microscopy). Ensure optimal imaging
settings (contrast, brightness, and resolution).
 Noise Reduction: Apply noise reduction techniques, such as Gaussian filtering,
bilateral filtering, or non-local means filtering, to enhance small structure visibility
while preserving edges.
 Normalization: Normalize the intensity values to standardize the range of pixel
values across the dataset. This helps ensure small structures are not overshadowed
by larger or brighter structures.

b. Segmentation
 Thresholding: Use adaptive or global thresholding to separate the structures of
interest from the background.
 Region-Growing: Identify and isolate connected regions corresponding to small
structures.
 Machine Learning: Use supervised or unsupervised machine learning models to
classify and segment small structures, leveraging tools like U-Net or similar deep
learning architectures for precise delineation.

c. Enhancement of Small Structures

 Contrast Enhancement: Apply histogram equalization or contrast-limited adaptive
histogram equalization (CLAHE) to improve the visibility of small structures by
increasing their contrast relative to the surrounding regions.
 Edge Enhancement: Use edge-detection filters like Sobel, Canny, or Laplacian filters
to highlight boundaries of small structures. Apply anisotropic diffusion to enhance
edges while smoothing noise.
 Morphological Operations: Use operations like dilation, erosion, or skeletonization to
emphasize the small structures without introducing artifacts.

d. Orthogonal Projections
Orthogonal projections involve projecting the 3D data onto 2D planes (XY, XZ, YZ) or
visualizing the data along specific orientations to highlight structures.
a. Choose the Projection Plane
Select planes of interest based on the orientation of the structures. For
example:
 XY for top-down views.
 XZ or YZ for side views.

b. Weight the Projections

Use intensity-weighted projections to highlight small structures:
 Maximum Intensity Projection (MIP): Projects the maximum intensity
along the viewing axis, enhancing bright structures.
 Minimum Intensity Projection (MinIP): Projects the minimum intensity,
useful for darker structures.
 Average Projection: Takes the mean intensity, reducing noise and
balancing bright and dark areas.

c. Multiscale Projection
Generate projections at multiple scales to visualize details at different levels of
magnification, providing both a macro and micro view of the structures.
e. Improved Visualization

 Color Mapping: Apply color maps to represent the intensity gradient, making small
structures more discernible. Use perceptually uniform color schemes (e.g., Viridis,
Cividis) to avoid misinterpretation.
 Transparency and Opacity Adjustments: Adjust transparency settings in overlapping
regions to make hidden small structures more visible in composite views.
 Annotation: Automatically annotate the small structures in the projection views with
labels, contours, or markers for better identification.

f. Integration with 3D Visualization

Combine orthogonal projections with 3D visualization:

 Use slicing tools to view orthogonal planes interactively.
 Overlay orthogonal projections on 3D renderings to provide context for small
structures.

6. How can surface models be used to visualize anatomical structures and identify
abnormalities?

Surface Models for Visualizing Anatomical Structures and Identifying Abnormalities

Surface models represent the boundary or outer layer of 3D anatomical structures derived from
medical imaging data. They are essential in medical visualization for providing a clear,
interactive, and intuitive representation of anatomy, aiding in diagnostics, treatment planning,
and surgical simulation.

a. Surface Models: Definition and Construction

A surface model is a 3D geometric representation of the outer surface of an object. In
medical imaging, these models are typically derived from imaging modalities such as CT,
MRI, or ultrasound.

Steps to Construct Surface Models:

 Image Acquisition: Obtain volumetric imaging data (e.g., CT or MRI scans) with high
spatial resolution.
 Segmentation: Delineate the region of interest (ROI) from the image volume using
segmentation techniques:
 Thresholding: Based on intensity values.
 Region Growing: Identifying connected regions with similar properties.
 Deep Learning Models: Automated segmentation using neural networks.
 Surface Extraction: Use algorithms like Marching Cubes or Level Sets to extract the
3D surface from the segmented ROI.
 Mesh Generation: Generate a polygonal mesh (e.g., triangular or quadrilateral) to
represent the surface.
 Smoothing and Refinement: Apply smoothing techniques (e.g., Laplacian smoothing)
to remove noise or irregularities in the mesh.
b. Applications of Surface Models in Visualization

 Representation of Anatomical Structures:

 Detailed Geometry: Surface models capture the shape, size, and spatial
relationships of structures such as bones, organs, or blood vessels.
 Interactive Visualization: Enables rotation, zooming, and slicing for detailed
examination of anatomical features.

 Identification of Abnormalities:

 Structural Deformities: Compare surface models of normal anatomy with

patient-specific models to identify deviations such as fractures, deformities, or
abnormal growths.
 Tumor Detection: Surface models can delineate tumor boundaries, showing
size, shape, and proximity to critical structures.
 Aneurysm Analysis: Visualize bulges in blood vessels and assess risk based
on size and shape irregularities.

c. Techniques for Identifying Abnormalities Using Surface Models

 Comparison with Reference Models: Overlay patient-specific surface models onto

standard anatomical models to identify discrepancies.
 Surface Color Mapping: Apply color coding to indicate abnormalities, such as:

 Regions of high curvature for fractures.

 Thickness variations for vascular diseases.

 Deformable Models: Use deformable surface models to fit patient-specific data and
highlight areas of discrepancy or abnormal deformation.
 Quantitative Analysis: Measure geometric parameters like volume, surface area, or
curvature to detect abnormalities (e.g., hypertrophy or atrophy in organs).

d. Use Cases of Surface Models in Identifying Abnormalities

 Orthopedics: Surface models of bones are used to visualize fractures, dislocations,

or bone density abnormalities. Models aid in preoperative planning for joint
replacement or fracture fixation.
 Neurology: Brain surface models highlight structural changes due to tumors, trauma,
or neurodegenerative diseases. Cortical thickness mapping helps diagnose
conditions like Alzheimer’s disease.
 Cardiovascular Imaging: Surface models of blood vessels identify abnormalities like
aneurysms, stenosis, or plaque buildup. Heart surface models visualize structural
defects or deformation in congenital heart diseases.
 Oncology: Tumor boundaries are visualized using surface models, enabling precise
measurement of size and interaction with surrounding tissues.
 Surgical Planning and Navigation: Preoperative surface models of organs and
vessels guide surgeons during complex procedures by highlighting critical structures
and abnormalities.

e. Advantages

 Intuitive Visualization
 Quantitative and Qualitative Analysis
  Interactivity
 Integration with Simulation
 Enhanced Diagnostic Accuracy

f. Challenges

 Image Quality Dependency

 Segmentation Errors
 Computational Complexity
 Limited Representation of Internal Features

7. How can volume rendering techniques be used to visualize 3D structures within

medical images?

Volume Rendering Techniques for Visualizing 3D Structures in Medical Images

Volume rendering is a powerful technique used to visualize 3D structures from medical imaging
data, such as CT, MRI, or PET scans. Unlike surface rendering, which focuses on external
boundaries, volume rendering allows for the visualization of internal structures, providing a
detailed and interactive view of the anatomy. Below is a comprehensive explanation of the
principles, techniques, and applications of volume rendering in medical imaging. Volume
rendering is a 3D visualization technique that processes volumetric data to create a 2D image.
It maps intensity values (voxels) from the imaging dataset into color and opacity to represent
the anatomical structures comprehensively.

a. Image Acquisition and Preprocessing

 Data Acquisition: Collect volumetric datasets from imaging modalities like CT or MRI.
 Pre-processing: Noise reduction, normalization, and contrast enhancement improve
the quality of visualization.

b. Transfer Functions

 Intensity Mapping: Map voxel intensity values to color and opacity using transfer
functions.

 Low intensities (e.g., air) → Transparent or black.

 Medium intensities (e.g., soft tissues) → Semi-transparent colors.

 High intensities (e.g., bones) → Opaque and bright colors.

 Customizable Functions: Clinicians can adjust transfer functions to highlight specific

structures (e.g., bones, vessels, or tumors).
c. Ray Casting

 A ray is cast from the viewer's perspective through the 3D volume.

 Along each ray: Voxel intensities are sampled. Colours and opacities are
accumulated to compute the final pixel colour.

d. Rendering Techniques

 Direct Volume Rendering (DVR):Directly visualizes the data without intermediate

geometric representation.

 Maximum Intensity Projection (MIP):Projects the voxel with the highest intensity
along the ray to the image plane.

 Iso-Surface Rendering:Displays surfaces at a specific intensity threshold (hybrid of

surface and volume rendering).

Applications of Volume Rendering in Medical Imaging

a. Anatomical Visualization

 Internal Organs: Visualize internal structures like the brain, lungs, heart, or liver with
detailed internal textures.
 Bone Structures: Highlight bones and joints for fracture analysis or surgical
planning.

b. Multimodal Image Fusion : Combine datasets from different modalities (e.g., CT and PET)
to visualize both anatomical and functional information in a single volume rendering.

c. Tumor and Lesion Analysis: Visualize and quantify tumors or lesions in organs, aiding in
diagnosis and treatment planning.

d. Vascular Imaging: Highlight blood vessels to assess conditions like aneurysms, stenosis,
or blockages.

e. Preoperative Planning: Provide surgeons with a detailed 3D view of the anatomy to plan
minimally invasive procedures or complex surgeries.

Advantages

 Comprehensive Visualization
 Customizable Views
 Non-Invasive Analysis
 Multiscale Representation

Challenges

 Computational Complexity
 Transfer Function Design
 Artifacts
 User Dependency
8. Evaluate the challenges associated with multimodal image registration. Discuss how
rigid, affine, and non-rigid transformations are applied to register CT, MRI, and PET
images, and demonstrate their implementation in MATLAB.

Challenges Associated with Multimodal Image Registration

Multimodal image registration aligns images from different imaging modalities (e.g., CT, MRI,
PET) to combine their complementary information for accurate diagnosis and treatment
planning. This process faces several challenges:

a. Differences in Image Characteristics: Contrast Variations: CT highlights bone structures,

MRI emphasizes soft tissues, and PET shows functional activity. Intensity Mismatch: No
direct correlation between intensity values in different modalities.

b. Noise and Artifacts: Noise from MRI or PET images and artifacts from patient movement
can hinder accurate alignment.

c. Nonlinear Deformations: Differences in patient positioning or organ deformation due to

breathing can cause mismatches.

d. Anatomical Complexity: Complex structures with varying shapes, sizes, and orientations
make alignment challenging.

e. Computational Demand: Multimodal registration, especially with non-rigid transformations,

is computationally intensive.

f. Lack of Ground Truth: Defining a "correct" alignment is often subjective and depends on the
clinical context.

Rigid, Affine, and Non-Rigid Transformations in Image Registration

These transformations align images by mapping points from one image (moving image) to
another (fixed image):

a. Rigid Transformation: Involves translation and rotation without altering shape or size.
Suitable for registering images of rigid structures (e.g., bones in CT and MRI).

b. Affine Transformation: Extends rigid transformations by including scaling and shearing.

Suitable for global alignment when scaling differences exist (e.g., between PET and MRI).

c. Non-Rigid Transformation: Applies local deformations, allowing flexibility to account for

nonlinear differences.

Common methods:

 Spline-based (e.g., B-splines): Models deformation using control points.

 Elastic or Demons Algorithm: Simulates physical deformation.

Application: Used for deformable structures like soft tissues or organs affected by
breathing.
Implementation of Multimodal Registration in MATLAB

Below is an example of registering CT, MRI, and PET images using rigid, affine, and non-rigid
transformations in MATLAB.

Load and Visualize Images

matlab
% Load images
ctImage = imread('CT_image.png');
mriImage = imread('MRI_image.png');
petImage = imread('PET_image.png');

% Convert to grayscale if necessary

ctImage = rgb2gray(ctImage);
mriImage = rgb2gray(mriImage);
petImage = rgb2gray(petImage);

% Visualize images
figure;
subplot(1,3,1), imshow(ctImage), title('CT Image');
subplot(1,3,2), imshow(mriImage), title('MRI Image');
subplot(1,3,3), imshow(petImage), title('PET Image');

Rigid Registration
matlab
% Select the fixed and moving images
fixedImage = ctImage;
movingImage = mriImage;

% Create a configuration for rigid transformation

[optimizer, metric] = imregconfig('monomodal');

% Perform rigid registration

rigidRegistered = imregister(movingImage, fixedImage, 'rigid', optimizer, metric);

% Display results
figure;
imshowpair(fixedImage, rigidRegistered, 'blend');
title('Rigid Registration: CT and MRI');

Affine Registration
matlab
% Configure and perform affine registration
affineRegistered = imregister(movingImage, fixedImage, 'affine', optimizer, metric);

% Display results
figure;
imshowpair(fixedImage, affineRegistered, 'blend');
title('Affine Registration: CT and MRI');
Non-Rigid Registration
matlab
% Create an object for non-rigid registration
deformationField = imregdemons(movingImage, fixedImage, [100 50 25],
'AccumulatedFieldSmoothing', 2);

% Apply the deformation field

nonRigidRegistered = imwarp(movingImage, deformationField);

% Display results
figure;
imshowpair(fixedImage, nonRigidRegistered, 'blend');
title('Non-Rigid Registration: CT and MRI');
Registration with PET Images

 Repeat the steps above with the PET image as the moving image and either CT or MRI
as the fixed image.

 Adjust optimizer and metric settings for multimodal registration:

Matlab

[optimizer, metric] = imregconfig('multimodal');

9. Discuss the role of 3D medical image visualization in personalized medicine. How do

surface-based and volume-based rendering contribute to treatment planning?
Provide a MATLAB-based approach to visualize a 3D model of a human organ.

3D Medical Image Visualization in Personalized Medicine

3D medical image visualization plays a crucial role in personalized medicine by providing

detailed, patient-specific insights into anatomical structures and pathologies. It aids in
diagnosis, treatment planning, surgical navigation, and monitoring therapeutic outcomes.

Role in Personalized Medicine

a. Patient-Specific Diagnosis: 3D visualization helps clinicians accurately analyze individual

anatomical variations and identify abnormalities like tumors, fractures, or vascular issues.
b. Treatment Planning: Personalized models allow precise planning of surgeries or therapies,
considering the patient’s unique anatomy. For example, reconstructive surgeries or
radiation therapy can be tailored using 3D organ models.
c. Surgical Guidance: 3D visualizations integrated with augmented reality (AR) or virtual
reality (VR) enable real-time surgical navigation.
d. Monitoring and Predictive Modeling: Compare pre- and post-treatment 3D models to
evaluate therapeutic effectiveness. Predict disease progression using dynamic models.
Surface-Based vs. Volume-Based Rendering in Treatment Planning

Surface-Based Rendering:

Focus: Displays the external boundaries or surfaces of structures.

Key Features:

 Fast rendering due to reduced computational demand.

 Useful for visualizing bones, organ boundaries, and surgical plans.

Applications:

 Orthopedic surgery: Bone structure visualization.

 Tumor boundary delineation.

 Implant design and placement.

Volume-Based Rendering:

Focus: Displays internal structures by rendering the entire volume of data.

Key Features:

 Provides a comprehensive view, including soft tissues and internal features.

 Uses transfer functions to map voxel intensities to color and opacity.

Applications:

 Visualizing tumors, blood vessels, or brain structures.

 Preoperative assessment of complex cases (e.g., vascular malformations).

 Radiation therapy planning.

Combined Approach:

 Integrating both methods allows surface rendering for anatomical boundaries and volume
rendering for internal features, offering a complete view.

MATLAB-Based Approach for 3D Organ Visualization

Below is a step-by-step MATLAB-based approach to visualize a 3D model of a human organ

using surface and volume rendering techniques.

Step 1: Load 3D Medical Data

matlab
% Load 3D medical imaging data
data = load('patient_CT.mat'); % Replace with actual file
volumeData = data.CT; % Volume matrix

% Display a slice for verification

figure;
imshow(volumeData(:,:,50), []); % Display the 50th slice
title('Slice of CT Data');

Step 2: Pre-process the Data

matlab
% Normalize the data
volumeData = mat2gray(volumeData);

% Apply a threshold to segment the organ of interest

threshold = 0.5; % Adjust based on intensity values
segmentedVolume = volumeData > threshold;

% Smooth the segmented volume

smoothedVolume = imgaussfilt3(double(segmentedVolume), 2);
Step 3: Surface Rendering
matlab
% Extract the surface using the Marching Cubes algorithm
fv = isosurface(smoothedVolume, 0.5); % 0.5 is the isovalue for the surface

% Visualize the surface

figure;
patch(fv, 'FaceColor', [0.8, 0.3, 0.3], 'EdgeColor', 'none');
camlight;
lighting gouraud;
title('Surface Rendering of the Organ');
axis equal;
Step 4: Volume Rendering
matlab
% Define a volume rendering view
figure;
volshow(volumeData, 'Renderer', 'MaximumIntensityProjection');
title('Volume Rendering of the Organ');
Step 5: Interactive 3D Visualization
matlab
% Combine surface and volume rendering
figure;

% Add volume rendering

hVolume = volshow(volumeData, 'Renderer', 'VolumeRendering');
hold on;

% Add surface rendering

hSurface = patch(fv, 'FaceColor', [0.8, 0.3, 0.3], 'EdgeColor', 'none');
hold off;

% Adjust lighting and view

lighting gouraud;
camlight;
title('Combined Surface and Volume Rendering');
Advantages:

 Precision: Provides accurate spatial relationships between structures.

 Comprehensiveness: Combines internal (volume) and external (surface) features.
 Interactivity: Allows clinicians to rotate, zoom, and explore patient-specific anatomy.

Challenges:

 Computational Complexity: Real-time rendering of large datasets requires high-

performance hardware.

 Data Quality: Noise and artifacts in imaging data can affect model accuracy.

 Transfer Function Design: Volume rendering depends on well-defined transfer

functions, which may require manual tuning.

10. What is the significance of principal component analysis (PCA) in image registration
and dimensionality reduction for medical images?

Significance of Principal Component Analysis (PCA) in Image Registration and Dimensionality

Reduction for Medical Images are:

Principal Component Analysis (PCA) is a widely used statistical method in medical imaging for
dimensionality reduction and feature extraction. It transforms high-dimensional data into a
lower-dimensional space while retaining most of the data's variability. PCA plays a crucial role
in image registration and dimensionality reduction, enhancing the efficiency and accuracy of
medical image analysis.

a. Role of PCA in Image Registration: Image registration aligns two or more images into a
common coordinate system. PCA aids this process in the following ways:
b. Feature Extraction: PCA identifies dominant features or patterns in images by analyzing
their variance. It extracts meaningful structures (e.g., organ boundaries) that can serve as
landmarks for registration.
Example: In brain imaging, PCA can extract principal axes of anatomical features like
ventricles or tumors, aiding alignment.
c. Preprocessing for Alignment: PCA reduces noise and irrelevant details, ensuring
registration algorithms focus on significant features. Aligning the principal components of
the fixed and moving images provides an initial estimate for registration.
d. Handling Multimodal Data: PCA helps identify common features in images from different
modalities (e.g., CT and MRI), enabling better alignment despite intensity differences.
e. Computational Efficiency: PCA reduces the dimensionality of the image data, speeding up
iterative optimization methods used in registration algorithms.

Role of PCA in Dimensionality Reduction

Dimensionality reduction is critical in medical imaging, as datasets are often high-dimensional,

with millions of voxels per scan. PCA simplifies the data while preserving its essential
structure.
 i. Reduction of Redundant Data : Medical images often contain spatial redundancy. PCA
reduces the data size by projecting it onto a smaller number of principal components,
making storage and processing more efficient.
ii. Noise Reduction: PCA separates signal from noise by capturing major variations in the
data. Small principal components with low variance often represent noise and can be
discarded.
iii. Visualization of High-Dimensional Data: PCA transforms multi-dimensional data into 2D or
3D for visualization, aiding in understanding complex structures.
iv. Accelerating Machine Learning Models: Dimensionality-reduced data reduces training time
for models like neural networks while improving generalization by eliminating irrelevant
features.

Applications of PCA in Medical Imaging

a. Image Compression: PCA compresses high-resolution images for efficient storage and
transmission without significant quality loss.

b. Tumor Detection: PCA isolates dominant features in images, enhancing tumor boundaries
for segmentation and diagnosis.

c. Multimodal Fusion: Combines features from different imaging modalities, such as CT and
MRI, into a common lower-dimensional space.

d. Shape Analysis: PCA is used in statistical shape models to analyze anatomical variations in
organs or tissues.

PCA for Dimensionality Reduction in MATLAB

Step 1: Load Image Data

matlab
% Load a 3D medical image
data = load('medical_image.mat'); % Replace with actual file
imageData = data.image;

% Reshape 3D data into 2D matrix (rows: voxels, columns: slices)

[rows, cols, slices] = size(imageData);
reshapedData = reshape(imageData, [], slices);
Step 2: Perform PCA
matlab
% Perform PCA
[coeff, score, ~, ~, explained] = pca(reshapedData);

% Determine the number of components to retain 95% variance

cumulativeVariance = cumsum(explained);
numComponents = find(cumulativeVariance >= 95, 1);

% Reduce the dimensionality

reducedData = score(:, 1:numComponents);
Step 3: Visualize Principal Components
matlab
% Reconstruct image using top components
reconstructedData = reducedData * coeff(:, 1:numComponents)';
reconstructedImage = reshape(reconstructedData, rows, cols, slices);
% Display original and reconstructed slices
figure;
subplot(1, 2, 1);
imshow(imageData(:, :, round(slices/2)), []);
title('Original Slice');

subplot(1, 2, 2);
imshow(reconstructedImage(:, :, round(slices/2)), []);
title('Reconstructed Slice with PCA');

Benefits of PCA in Medical Imaging

 Efficiency: Reduces data size, speeding up processing and analysis.

 Noise Robustness: Filters out noise by focusing on significant variance.
 Interpretability: Simplifies complex data, aiding visualization and feature selection.

Limitations of PCA in Medical Imaging

 Linear Assumption: PCA assumes linear relationships, which may not hold for complex
medical images.
 Loss of Information: Small principal components may still carry useful details.
 Scaling Dependency: PCA requires proper normalization of input data.