CEP

Submitted By: Bazia Azmat

Submitted to: Ma’am Shahinza Manzoor

Roll Number: 1016

Program: BSCE-6

Course Title: DSP (Lab)

Page 1 of 15
Medical imaging system is being developed to aid in the detection of
breast cancer. The system uses digital signal processing techniques to
analyze images obtained from mammography scans and employs
machine learning algorithms to automatically identify regions of
interest that may contain tumors. The images are first preprocessed to
enhance contrast and remove noise, and then segmented to separate
the breast tissue from the background. Next, feature extraction
techniques are used to extract morphological, texture, and density
features from the segmented images, which are used as inputs to the
machine learning algorithms. The machine learning algorithms
include supervised learning algorithms such as support vector
machines (SVMs), decision trees, and random forests, as well as
unsupervised learning algorithms such as k-means clustering and
principal component analysis (PCA). Question

1: Trace some of the preprocessing steps used in the medical imaging system
to enhance mammography images and remove noise?

In the medical imaging system designed to aid in the detection of breast cancer,
several preprocessing steps are employed to enhance mammography images and
remove noise. Some of these preprocessing techniques include:

Image Denoising: Mammography images often contain noise, which can interfere
with the accurate identification of potential tumors. Denoising techniques such as

Page 2 of 15
filtering are used to reduce the noise and improve image quality. Common filtering
methods include median filtering, Gaussian filtering, or adaptive filtering.

Contrast Enhancement: Enhancing the contrast of mammography images helps to

improve the visibility of subtle details, including small tumors or
microcalcifications. Techniques like histogram equalization, contrast stretching, or
adaptive contrast enhancement are applied to adjust the image’s dynamic range and
increase the visual contrast.

Image Calibration: It is crucial to calibrate mammography images to ensure

consistent and accurate intensity values across different images. Calibration
corrects for variations in X-ray beam intensity, detector response, and other
imaging system characteristics. This step helps in normalizing the images and
improving the accuracy of subsequent analysis.

Image Registration: In some cases, multiple mammography images of the same

breast are taken at different angles or time points. Image registration techniques
align these images spatially, allowing for a comprehensive examination and
comparison. Registration methods may involve rigid or deformable
transformations to align the images accurately.

Artifact Removal: Mammography images can contain artifacts due to various

factors, such as motion during image acquisition or compression artifacts.

Page 3 of 15
Preprocessing techniques are used to identify and remove these artifacts, ensuring
that they do not interfere with the subsequent analysis.

Image Rescaling and Resampling: Mammography images may have different

resolutions or dimensions. Rescaling and resampling techniques are used to ensure
uniformity in image size and resolution, which facilitates consistent analysis and
comparison.

These preprocessing steps collectively contribute to improving the quality, clarity,

and accuracy of mammography images, enabling more reliable tumor detection
and analysis in the subsequent stages of the medical imaging system.

Page 4 of 15
Question 2: Explain how the images are segmented to separate the breast
tissue from the background, and why is this step important for detecting
tumors.

Segmentation plays a crucial role in separating the breast tissue from the
background in mammography images. This step is important for detecting tumors
because it allows the focus of analysis to be directed specifically towards the breast
region, where tumors are most likely to occur. Here’s an explanation of how
segmentation is typically performed in this context:

Thresholding: One common technique used for segmentation is thresholding,

where a specific threshold value is chosen to distinguish between the foreground

Page 5 of 15
(breast tissue) and the background. Pixels with intensity values above the threshold
are classified as foreground, while those below the threshold are classified as
background. This method assumes that the intensities of the breast tissue and
background have distinct characteristics.

Region Growing: Region growing techniques involve selecting seed points in the
breast region and iteratively expanding the region by adding neighboring pixels
that meet certain criteria. The criteria can be based on intensity similarity, texture
features, or other characteristics specific to breast tissue. Region growing methods
exploit the assumption that the breast region is a connected and homogeneous area.

Active Contour Models: Active contour models, also known as snakes, are
deformable curves or surfaces that are iteratively adjusted to fit the boundaries of
the breast tissue. These models are initialized close to the breast boundary and are
guided by forces such as gradient, curvature, and external constraints. Active
contour models can effectively capture irregular and complex shapes, making them
useful for breast segmentation.

Machine Learning-Based Segmentation: Machine learning techniques can also be

employed for segmentation, where algorithms are trained to classify pixels or
regions as breast tissue or background. This approach utilizes labeled training data
to learn the characteristics and patterns that distinguish the breast region. Common
machine learning algorithms used for segmentation include convolutional neural
networks (CNNs) and support vector machines (SVMs).

Page 6 of 15
The segmentation step is crucial for tumor detection because it isolates the breast
tissue, reducing the complexity and variability of the image data. By focusing the
analysis on the breast region, the algorithms can specifically target and identify
potential tumors within the segmented area. Segmentation enables accurate
localization, size estimation, and characterization of tumors, facilitating subsequent
stages of feature extraction and machine learning algorithms that aid in tumor
identification and classification.

Page 7 of 15
Page 8 of 15
Question 3: Show some of the feature extraction techniques used in the
medical imaging system to extract morphological, texture, and density
features from the segmented images.

In the medical imaging system designed for breast cancer detection, various feature
extraction techniques are employed to extract morphological, texture, and density
features from the segmented images. These features provide valuable information
about the characteristics and properties of potential tumors. Here are some
commonly used feature extraction techniques:

Page 9 of 15
Morphological Features:

Area: The total number of pixels within a segmented region, which provides an
estimate of the tumor size.

Perimeter: The length of the boundary of the segmented region, which indicates the
shape irregularity.

Compactness: A measure of how closely the tumor is packed within its boundary,
calculated as the ratio of perimeter squared to the area.

Eccentricity: A measure of the elongation of the tumor, ranging from 0 (perfect

circle) to 1 (line).

Convexity: The ratio of the area of the segmented region to its convex hull,
indicating the extent of concavity or convexity.

Texture Features:

Gray-Level Co-occurrence Matrix (GLCM): GLCM measures the statistical

relationships between pairs of pixels at different distances and orientations.
Features extracted from GLCM include energy, entropy, contrast, homogeneity,
and correlation, which characterize the texture properties of the segmented region.

Gray-Level Run Length Matrix (GLRLM): GLRLM analyzes the lengths and
frequencies of runs of pixels with the same intensity value. Features extracted from
GLRLM include short-run emphasis, long-run emphasis, gray-level nonuniformity,
and run-length nonuniformity, providing information about the texture complexity
and homogeneity.

Local Binary Patterns (LBP): LBP encodes the relationship between a pixel and its
neighbors by comparing their intensity values. Histograms of LBP patterns are
computed within the segmented region to capture texture patterns and variations.
Page 10 of 15
Density Features:

Density Histogram: A histogram is created based on the pixel intensities within the
segmented region, representing the distribution of pixel densities. Features such as
mean, standard deviation, skewness, and kurtosis of the histogram provide
information about the density characteristics of the region.

Histogram of Oriented Gradients (HOG): HOG computes the distribution of

gradient orientations within the segmented region. It captures the edge and gradient
information, which can be indicative of tumor boundaries and shape.

These feature extraction techniques aim to capture different aspects of the

segmented region, including shape, texture patterns, and density characteristics.
The extracted features serve as inputs to machine learning algorithms, enabling the
system to learn patterns and make accurate predictions regarding the presence or
absence of breast tumors.

Question 4: Extracted features used as inputs to the machine learning

algorithms, and shows some examples of the supervised and unsupervised
learning algorithms used in the system?

Once the features are extracted from the segmented images, they are used as inputs
to machine learning algorithms for further analysis and classification. Here are
examples of both supervised and unsupervised learning algorithms commonly used
in the medical imaging system for breast cancer detection:

Supervised Learning Algorithms:

Support Vector Machines (SVM): SVM is a popular algorithm used for

classification tasks. It constructs a hyperplane that maximally separates different
classes based on the extracted features. SVM can handle both linear and nonlinear

Page 11 of 15
classification problems and has been successfully applied in medical imaging for
tumor detection and classification.

Decision Trees: Decision trees partition the feature space based on a series of if-
then rules. They are capable of handling both categorical and continuous features
and are interpretable. Decision trees can be used for binary or multi-class
classification tasks and have been widely used in medical imaging for breast
cancer detection.

Random Forests: Random forests are an ensemble learning method that combines
multiple decision trees. Each tree in the forest is trained on a random subset of the
training data and features. Random forests are effective in reducing overfitting and
improving classification accuracy. They are robust against noise and outliers and
have been used in various medical imaging applications, including breast cancer
detection.

Unsupervised Learning Algorithms:

K-means Clustering: K-means is a popular clustering algorithm used to group

similar data points based on their feature similarities. In the medical imaging
system, K-means clustering can be applied to segment the data into different
groups or clusters, potentially identifying subtypes of breast tumors or grouping
similar mammography images together.

Principal Component Analysis (PCA): PCA is a dimensionality reduction

technique that transforms the high-dimensional feature space into a lower-
dimensional space while preserving the most important information. In the context
of breast cancer detection, PCA can help in reducing the feature dimensionality
and identifying the most discriminative features for tumor classification.

Page 12 of 15
Hierarchical Clustering: Hierarchical clustering is a method that builds a hierarchy
of clusters by recursively merging or splitting them. It can be used to identify
clusters or patterns in the breast cancer dataset based on the extracted features.
Hierarchical clustering can aid in understanding the relationships between different
types of breast tumors or grouping similar cases together.

These machine learning algorithms, both supervised and unsupervised, leverage

the extracted features to learn patterns, classify tumors, and assist in the accurate
detection and diagnosis of breast cancer. The choice of algorithm depends on the
specific requirements, dataset characteristics, and the nature of the problem being
addressed in the medical imaging system.

Question 5: States how is the performance of the medical imaging system

evaluated, and what are some metrics used to assess the accuracy of the
system in detecting breast cancer?

The performance of the medical imaging system for breast cancer detection is
typically evaluated using various metrics to assess its accuracy and effectiveness.
Here are some commonly used metrics for evaluating the performance of the
system:

Sensitivity: Sensitivity, also known as the true positive rate or recall, measures the
proportion of actual positive cases (breast cancer) correctly identified by the
system. It indicates the system’s ability to detect true positive cases without
missing them.

Specificity: Specificity measures the proportion of actual negative cases (no breast
cancer) correctly identified by the system. It indicates the system’s ability to
correctly classify negative cases, minimizing false positives.

Page 13 of 15
Accuracy: Accuracy measures the overall correctness of the system’s predictions
by considering both true positives and true negatives. It calculates the percentage
of correctly classified cases out of the total number of cases.

Precision: Precision measures the proportion of correctly identified positive cases

out of all the cases predicted as positive by the system. It quantifies the system’s
ability to avoid false positives.

F1 Score: The F1 score combines precision and sensitivity into a single metric,
providing a balance between the two. It is the harmonic mean of precision and
sensitivity and is useful when there is an uneven distribution between positive and
negative cases.

Receiver Operating Characteristic (ROC) Curve: The ROC curve plots the true
positive rate (sensitivity) against the false positive rate (1 – specificity) for
different classification thresholds. It provides a visual representation of the trade-
off between sensitivity and specificity, and the area under the ROC curve (AUC) is
often used as a summary metric of the system's performance.

Confusion Matrix: A confusion matrix provides a tabular representation of the

system’s performance, showing the counts of true positives, true negatives, false
positives, and false negatives. It allows for a more detailed analysis of the system’s
classification performance and is useful for calculating metrics such as sensitivity,
specificity, and accuracy.

Cross-Validation: Cross-validation is a technique used to assess the generalization

performance of the system by splitting the dataset into multiple subsets. The
system is trained and evaluated on different subsets, and the average performance
across the subsets is calculated. Common cross-validation techniques include k-
fold cross-validation and leave-one-out cross-validation.

Page 14 of 15
These metrics and evaluation techniques help in quantitatively assessing the
accuracy, reliability, and performance of the medical imaging system for breast
cancer detection. They provide valuable insights into the system's ability to
correctly identify tumors and classify breast cancer cases, allowing for refinement
and improvement of the system’s algorithms and techniques.

Page 15 of 15