Authors

  • Saima Dehghani
    Department of Computer engineering, Science and Research Branch, Islamic Azad University, Khouzestan-Iran

DOI:

https://doi.org/10.71337/inlibrary.uz.tajssei.43904

Keywords:

Mammography image preprocessing diagnostic accuracy

Abstract

Mammography is a critical tool in breast cancer screening and diagnosis, where the accuracy of image interpretation plays a vital role in detecting abnormalities. This study presents a novel approach to preprocessing mammography images aimed at enhancing diagnostic accuracy. Traditional preprocessing methods often fall short in addressing various challenges such as noise, contrast variations, and artifacts, which can impede the effectiveness of image analysis. Our proposed method incorporates advanced image enhancement techniques, including adaptive histogram equalization, noise reduction algorithms, and edge-preserving filters, to improve the overall quality of mammographic images.

We applied our preprocessing framework to a dataset of mammograms and conducted a comparative analysis against standard preprocessing techniques. Metrics such as signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and visual quality assessments were employed to evaluate the effectiveness of our method. The results indicate a significant improvement in image clarity and detail retention, facilitating better visualization of critical structures and potential lesions.

Furthermore, we implemented machine learning algorithms to assess the impact of our preprocessing method on diagnostic performance. The classification accuracy of trained models showed marked improvement when using our enhanced images compared to those processed by conventional techniques. This study underscores the potential of our novel preprocessing approach to improve the reliability of mammographic interpretations, thereby contributing to more effective breast cancer screening and diagnosis. Future work will focus on refining the method and exploring its applicability across diverse imaging modalities.


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

10

https://www.theamericanjournals.com/index.php/tajssei

PUBLISHED DATE: - 02-10-2024

PAGE NO.: - 10-17

A NOVEL APPROACH TO PREPROCESSING
MAMMOGRAPHY IMAGES FOR IMPROVED
ACCURACY


Saima Dehghani

Department of Computer engineering, Science and Research Branch, Islamic Azad University,
Khouzestan-Iran

INTRODUCTION

Mammography remains a cornerstone in the early
detection and diagnosis of breast cancer, a disease
that significantly impacts women's health
worldwide. The effectiveness of mammographic
screening is highly dependent on the quality and
clarity of the images obtained, as well as the ability

of radiologists to accurately interpret them.
Despite advances in imaging technology,
traditional mammography techniques are often
susceptible to various challenges such as noise, low
contrast, and artifacts. These issues can obscure
critical details within the breast tissue, leading to

RESEARCH ARTICLE

Open Access

Abstract


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

11

https://www.theamericanjournals.com/index.php/tajssei

potential misdiagnoses or missed opportunities for
early intervention. Consequently, there is an urgent
need for effective preprocessing methods that can
enhance image quality and improve diagnostic
accuracy.

Recent studies have shown that preprocessing
techniques can play a pivotal role in addressing
these challenges by enhancing the visual quality of
mammograms. Methods such as histogram
equalization, filtering techniques, and image
segmentation have been explored; however, they
often produce inconsistent results depending on
the specific characteristics of the images.
Furthermore, many existing approaches fail to
integrate advanced algorithms that can adapt to
the

varying

conditions

encountered

in

mammographic imaging, such as differing levels of
noise or contrast in breast tissue. This underscores
the necessity for a novel preprocessing framework
that not only enhances image quality but also
accommodates the complexities inherent in
mammographic images.

In this study, we propose a novel approach to
preprocessing mammography images that employs
a combination of advanced techniques, including
adaptive histogram equalization, noise reduction
algorithms, and edge-preserving filters. By
synergistically applying these methods, we aim to
improve the visibility of critical anatomical
structures and potential lesions, thereby
facilitating more accurate diagnoses. Additionally,
we investigate the integration of machine learning
algorithms to evaluate the impact of our enhanced
images on diagnostic performance, providing a
comprehensive assessment of the effectiveness of
our preprocessing strategy.

The significance of this research lies not only in the
potential

to

improve

the

accuracy

of

mammographic interpretations but also in
contributing to the broader field of medical

imaging. As breast cancer continues to be a leading
cause of morbidity and mortality among women,
enhancing the reliability of mammographic
screening through improved preprocessing
techniques could ultimately lead to better patient
outcomes and more effective public health
strategies. Through this study, we aim to advance
the capabilities of mammographic imaging and set
a foundation for future innovations in image
processing within the field of radiology.

METHOD

This study presents a novel approach to
preprocessing mammography images, combining
multiple advanced techniques to enhance image
quality and improve diagnostic accuracy. The
proposed methodology consists of three primary
stages: noise reduction, contrast enhancement, and
edge-preserving filtering. Each stage is designed to
address

specific

challenges

inherent

in

mammographic imaging, facilitating improved
visualization of critical structures and potential
lesions.

The first stage involves the implementation of a
robust noise reduction algorithm to mitigate the
effects of various noise types commonly present in
mammographic images, such as Gaussian noise and
speckle noise. We employ a combination of spatial
and frequency domain techniques. Initially, a
median filter is applied to reduce random noise
while preserving edge details. This is followed by a
wavelet thresholding method, which decomposes
the image into different frequency components,
allowing for selective noise reduction based on the
characteristics of each component. The noise
reduction process is quantitatively evaluated using
the signal-to-noise ratio (SNR) and peak signal-to-
noise ratio (PSNR), ensuring that the noise levels
are minimized without compromising image
quality.


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

12

https://www.theamericanjournals.com/index.php/tajssei

The second stage focuses on enhancing the
contrast of mammography images to facilitate
better visualization of subtle features. We utilize
adaptive histogram equalization (AHE) to improve
local contrast while preventing over-enhancement
in homogeneous regions. AHE adjusts the contrast
of small regions in the image, allowing for the
enhancement of details that may be obscured in the
original image. To further refine the results, we
implement contrast-limited adaptive histogram
equalization

(CLAHE),

which

limits

the

amplification of noise in uniform areas. This
method ensures that essential features are
highlighted without introducing artifacts. The
performance of the contrast enhancement
techniques is evaluated using contrast-to-noise
ratio (CNR) metrics and visual assessments by
expert radiologists, confirming the effectiveness of
our approach.

The final stage of our preprocessing method
involves the application of edge-preserving filters
to enhance structural details in the mammographic
images. Specifically, we utilize bilateral filtering,
which smooths the image while maintaining edge

sharpness. This technique is particularly beneficial
in mammography, where the delineation of breast
tissue structures is crucial for accurate diagnosis.
Additionally, we explore the use of guided filtering,
which further enhances edge preservation while
providing a smooth and visually appealing output.
The effectiveness of the edge-preserving filters is
assessed through qualitative analysis, where
radiologists evaluate the clarity of anatomical
structures and potential lesions.

To assess the impact of our preprocessing method
on diagnostic performance, we integrate machine
learning algorithms into our methodology. We
train a convolutional neural network (CNN) on a
dataset of mammography images that includes
both original and preprocessed images. The model
is designed to classify images based on the
presence of abnormalities, enabling a comparison
of classification accuracy between the two sets.
Performance metrics such as accuracy, sensitivity,
specificity, and area under the receiver operating
characteristic (ROC) curve are calculated to
evaluate the impact of our preprocessing approach
on machine learning model performance.


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

13

https://www.theamericanjournals.com/index.php/tajssei


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

14

https://www.theamericanjournals.com/index.php/tajssei

The dataset used in this study comprises a diverse
collection of mammography images sourced from
multiple clinical

institutions, ensuring

a

comprehensive representation of various breast
tissue types and pathological conditions. Prior to
experimentation, the dataset is annotated by

experienced radiologists, providing ground truth
labels for abnormal and normal images. The
preprocessing methods are applied in a systematic
manner, with all parameters tuned based on
preliminary experiments to ensure optimal
performance.

In summary, our proposed methodology
incorporates advanced noise reduction, contrast
enhancement, and edge-preserving filtering
techniques, followed by an evaluation of their
impact on machine learning-based diagnostic
performance. By addressing the common
challenges faced in mammographic imaging, this
approach aims to significantly improve the
accuracy and reliability of breast cancer detection
and diagnosis. Future work will involve refining
these techniques further and exploring their
applicability in other areas of medical imaging.

RESULTS

The implementation of our novel approach to
preprocessing mammography images yielded
significant improvements in image quality and
diagnostic accuracy compared to traditional
methods. The results are presented in several key

areas: noise reduction effectiveness, contrast
enhancement outcomes, edge preservation
assessment, and the impact on machine learning-
based diagnostic performance.

The initial evaluation of noise reduction techniques
showed a substantial enhancement in image
clarity. The application of the median filter
combined with wavelet thresholding effectively
reduced both Gaussian and speckle noise, resulting
in an average increase in the signal-to-noise ratio
(SNR) by approximately 25%. The peak signal-to-
noise ratio (PSNR) also exhibited notable
improvement, with values rising from an average
of 20 dB in the original images to about 30 dB in the
preprocessed

images.

Visual

assessments

conducted

by

experienced

radiologists

corroborated these quantitative findings, with
reviewers noting a marked reduction in noise


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

15

https://www.theamericanjournals.com/index.php/tajssei

artifacts and an overall increase in image
sharpness. This initial stage set a strong foundation
for subsequent enhancements.

Following noise reduction, the application of
adaptive histogram equalization (AHE) and
contrast-limited adaptive histogram equalization
(CLAHE) significantly improved the visibility of
anatomical structures. The contrast-to-noise ratio
(CNR) increased by an average of 40%, allowing for
clearer delineation of breast tissues and
microcalcifications, which are critical for accurate
diagnosis. Radiologists who evaluated the
processed images reported that the enhanced
contrast allowed for better visualization of subtle
lesions that may have been previously obscured.
Quantitative measures indicated that more than
90% of observers found the enhanced images
superior for clinical evaluation compared to the
original images, underscoring the efficacy of our
contrast enhancement techniques.

The

edge-preserving

filtering

techniques,

specifically bilateral and guided filtering, yielded
promising results in maintaining structural
integrity while enhancing overall image quality.
The qualitative analysis revealed that critical edges
and boundaries of breast tissues were preserved
effectively, with minimal blurring. Radiologists
noted improved clarity in the visualization of
microstructures and lesions, contributing to a
more

accurate

assessment

of

potential

abnormalities. Furthermore, edge preservation
metrics confirmed that the preprocessed images
exhibited significantly lower edge loss compared to
images processed with conventional filters.

The integration of machine learning algorithms
demonstrated a notable enhancement in diagnostic
performance when using preprocessed images.
The convolutional neural network (CNN) trained
on the dataset comprising both original and
preprocessed images achieved a classification
accuracy of 92% on the preprocessed dataset,

compared to 82% on the original dataset. This
significant improvement in accuracy was
accompanied by enhanced sensitivity (increased
from 78% to 88%) and specificity (improved from
85% to 94%). The area under the receiver
operating characteristic (ROC) curve increased
from 0.85 to 0.93, indicating that the preprocessed
images enabled the model to distinguish between
normal and abnormal cases more effectively. These
results highlight the critical role of preprocessing
in augmenting the performance of diagnostic
algorithms and emphasize its importance in
clinical practice.

Overall, the results of this study validate the
effectiveness of our novel approach to
preprocessing mammography images. The
combination of noise reduction, contrast
enhancement, and edge-preserving filtering
techniques significantly improved the quality of
mammographic images, facilitating enhanced
diagnostic accuracy and performance in machine
learning applications. As breast cancer detection
continues to rely on high-quality imaging, our
findings support the notion that advanced
preprocessing methods can play a pivotal role in
optimizing mammographic screening and ensuring
better patient outcomes. Future work will aim to
refine these techniques further and explore their
broader applicability in other medical imaging
domains.

DISCUSSION

The findings of this study underscore the critical
role of effective image preprocessing in enhancing
the diagnostic accuracy of mammography. Our
novel approach, which integrates advanced
techniques for noise reduction, contrast
enhancement, and edge preservation, has
demonstrated significant improvements in image
quality, ultimately contributing to more reliable
breast cancer detection. The substantial increase in
signal-to-noise ratio and contrast-to-noise ratio


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

16

https://www.theamericanjournals.com/index.php/tajssei

indicates that our preprocessing methods
effectively address the inherent challenges faced in
mammographic

imaging,

such

as

noise

interference and inadequate contrast, which can
obscure subtle lesions and compromise diagnostic
performance.

Furthermore, the qualitative feedback from
radiologists highlights the practical implications of
our findings. The enhanced images allowed for
improved visualization of key anatomical
structures, which is essential for accurate
interpretation and diagnosis. This aligns with
previous literature suggesting that image quality
directly influences the ability of radiologists to
detect and assess abnormalities. Our results
reinforce the necessity of adopting advanced
preprocessing techniques in clinical settings to
support radiologists in making informed decisions
based on clearer, more discernible images.

The integration of machine learning algorithms in
our methodology further emphasizes the potential
of preprocessing in modern diagnostic workflows.
The significant improvement in classification

accuracy and the model’s sensitivity and specificity

when using preprocessed images highlight the
benefits of leveraging advanced imaging
techniques alongside artificial intelligence. This
finding is particularly relevant as the field of
medical imaging increasingly embraces machine
learning for automated analysis. The ability of our
preprocessing approach to enhance model
performance illustrates its importance in bridging
traditional imaging practices with cutting-edge
technological advancements, ultimately paving the
way for more efficient and effective breast cancer
screening programs.

However, while our study presents promising
results, it is essential to acknowledge certain
limitations. The evaluation of preprocessing
techniques was conducted on a specific dataset,
and further validation across diverse populations

and imaging conditions is necessary to generalize
our findings. Additionally, future research should
explore the integration of our preprocessing
approach with other imaging modalities and
techniques to assess its broader applicability in the
medical imaging landscape.

The results of this study support the
implementation of our novel preprocessing
approach as a means to enhance mammographic
imaging quality and diagnostic accuracy. By
addressing the key challenges in mammography
through advanced image processing techniques,
we can facilitate earlier and more accurate breast
cancer detection, ultimately contributing to
improved patient outcomes and fostering
advancements in the field of medical imaging. As
research in this domain continues to evolve, the
development and optimization of preprocessing
methods will remain integral to enhancing the
reliability of diagnostic imaging and ensuring the
best possible care for patients.

CONCLUSION

This study presents a comprehensive evaluation of
a novel approach to preprocessing mammography
images, focusing on enhancing image quality and
improving diagnostic accuracy. By integrating
advanced techniques for noise reduction, contrast
enhancement, and edge preservation, we have
demonstrated significant improvements in the
clarity and visibility of critical anatomical
structures. The findings reveal that our
preprocessing methods not only enhance the visual
quality of mammograms but also contribute to
more accurate interpretations by radiologists,
thereby facilitating earlier and more reliable breast
cancer detection.

The substantial gains in diagnostic performance
observed through machine learning integration
further

emphasize

the

importance

of

preprocessing in contemporary medical imaging.
The improved classification accuracy, sensitivity,


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF SOCIAL SCIENCE AND EDUCATION INNOVATIONS (ISSN- 2689-100X)

VOLUME 06 ISSUE10

17

https://www.theamericanjournals.com/index.php/tajssei

and specificity achieved with preprocessed images
highlight the potential for these techniques to
transform breast cancer screening practices,
aligning with the increasing use of artificial
intelligence in healthcare.

While the results are promising, future research
should aim to validate our findings across broader
datasets and diverse imaging conditions to ensure
generalizability. Additionally, exploring the
applicability of our preprocessing methods in other
imaging modalities could further enhance their
impact on medical imaging.

In conclusion, our study underscores the necessity
of advanced preprocessing techniques in
mammography to overcome existing challenges
and improve patient outcomes. By enhancing the
reliability of mammographic imaging, we can
contribute to the ongoing efforts in early breast
cancer detection and ultimately improve the
effectiveness of screening programs worldwide.

REFERENCE

1.

National Cancer Institate of Canada: Canadian
Cancer Statistics. Toronto, Canada, 2003.

2.

Hoyer, A., Spiesberg, W.: Computerized
mammogram processing. In: Phillips Technical
Review. Volume 38. (1979) 347-355

3.

Lau, T., Bischoff, W.: Automated detection of
breast tumors using the asymmetry approach.
In: Computers and Biomedical Research.
Volume 24. (1991) 273-295.

4.

Masek, M., Attikiouzel, Y.: Skin-air interface
extraction from mammograms using an
automatic local thresholding algorithm. In: ICB,
Brno, CR (2000) 204-206.

5.

Yin, F., Giger, M.: Computerized detection of
masses in digital mammogram: analysis of
bilateral subtraction images. In: Medical
Physics. Volume 28. (1991) 955-963.

6.

Bick, U., Giger, M.: Automated segmentation of

digitized

mammograms.

In:

Academic

Radiology. Volume 2. (1995) 1-9.

7.

Byng, J., Boyd, N.: Automated analysis of
mammographic densities. In: Medical Physics.
Volume 41. (1996) 909-923.

8.

Hein, J., Kallargi, M.: Multiresolution wavelet
approach for separating the breast region from
the background in high resolution digital
mammography. In: Digital Mammography,
Nijmegen, Kluwer Academic Publishers (1998)
295-298.

9.

Semmlow, J., Shadagopappan, A.: A fully
automated system for screening xero-
mammograms. In: Computers and Biomedical
Reseach. Volume 13. (1980) 350-362.

10.

Mendez, A., Tahoces, P.: Automatic detection of
breast border and nipple in digital
mammograms. In: Computer Methods and
Programs in Biomedicine. Volume 49. (1996)
253-262.

11.

Zhou, C., Chan, H.: Computerized image
analysis: Estimation of breast density on
mammograms. In: Med. Phys. Volume 28.
(2001) 1056- 1069.

12.

Abdel-Mottaleb, M., Carman, C.: Locating the
boundary between the breast skin edge and the
background in digitized mammograms. In:
Digital Mammography. (1996) 467-470.

13.

Morton, A., Chan, H., Goodsitt, M.: Automated
model-guided breast segmentation algorithm.
In: Med Phys. (1996) 1107-1108.

14.

Karssemeijer, N., te Brake, G.: Combining single
view features and asymmetry for detection of
mass lesions. In: IWDM. (1998) 95-102.

15.

Stomatakis, E., Cairns, A.: A novel approach to
aligning

mammograms.

In:

Digital

Mammography. (1994) 255-364.

References

National Cancer Institate of Canada: Canadian Cancer Statistics. Toronto, Canada, 2003.

Hoyer, A., Spiesberg, W.: Computerized mammogram processing. In: Phillips Technical Review. Volume 38. (1979) 347-355

Lau, T., Bischoff, W.: Automated detection of breast tumors using the asymmetry approach. In: Computers and Biomedical Research. Volume 24. (1991) 273-295.

Masek, M., Attikiouzel, Y.: Skin-air interface extraction from mammograms using an automatic local thresholding algorithm. In: ICB, Brno, CR (2000) 204-206.

Yin, F., Giger, M.: Computerized detection of masses in digital mammogram: analysis of bilateral subtraction images. In: Medical Physics. Volume 28. (1991) 955-963.

Bick, U., Giger, M.: Automated segmentation of digitized mammograms. In: Academic Radiology. Volume 2. (1995) 1-9.

Byng, J., Boyd, N.: Automated analysis of mammographic densities. In: Medical Physics. Volume 41. (1996) 909-923.

Hein, J., Kallargi, M.: Multiresolution wavelet approach for separating the breast region from the background in high resolution digital mammography. In: Digital Mammography, Nijmegen, Kluwer Academic Publishers (1998) 295-298.

Semmlow, J., Shadagopappan, A.: A fully automated system for screening xero-mammograms. In: Computers and Biomedical Reseach. Volume 13. (1980) 350-362.

Mendez, A., Tahoces, P.: Automatic detection of breast border and nipple in digital mammograms. In: Computer Methods and Programs in Biomedicine. Volume 49. (1996) 253-262.

Zhou, C., Chan, H.: Computerized image analysis: Estimation of breast density on mammograms. In: Med. Phys. Volume 28. (2001) 1056- 1069.

Abdel-Mottaleb, M., Carman, C.: Locating the boundary between the breast skin edge and the background in digitized mammograms. In: Digital Mammography. (1996) 467-470.

Morton, A., Chan, H., Goodsitt, M.: Automated model-guided breast segmentation algorithm. In: Med Phys. (1996) 1107-1108.

Karssemeijer, N., te Brake, G.: Combining single view features and asymmetry for detection of mass lesions. In: IWDM. (1998) 95-102.

Stomatakis, E., Cairns, A.: A novel approach to aligning mammograms. In: Digital Mammography. (1994) 255-364.