A Comparative Analysis of SVM and CNN For Scoliosis Detection in Backbone X-Ray Images

A Comparative Analysis of SVM
and CNN for Scoliosis Detection

in Backbone X-ray Images
AFFILIATION: BINA NUSANTARA

UNIVERSITY
Authors
1 2 3
Joysun Wisesa Albert Geoffrey Kevin
Gunawan Gunawan Laurensius
4 5
Said Jurike V
Achmad Moniaga
Overview
Introduction
Literature Review
Methodology
Results &
Analysis
Conclusions
Introduction
Scoliosis is a medical condition that affects the curvature of a person’s spine.
Since it could lead to severe health issues, it is something that has to be dealt
with as quickly as possible. An early detection can contribute to this matter.
However, it is still undeniable that early detection still proves to be difficult.

With the complexity of the scoliosis patterns, they may look indistinguishable
and could be missed. To solve this problem, some approaches with
machine learning techniques have been used to enhance the accuracy of the
detection for scoliosis, and therefore an immediate treatment can be given to
it.
This research uses two prominent machine learning models, SVM and CNN
to conduct a comprehensive comparative analysis in the context of
scoliosis detection using X-ray images of the backbone.
Support Vector Machines (SVM) have been widely used in medical image
analysis for classification tasks. SVM excels in identifying patterns within
datasets and has demonstrated success in various medical applications. Its
application in scoliosis detection warrants exploration and evaluation.
Convolutional Neural Networks (CNN), a type of deep learning model, have

shown remarkable success in image recognition tasks. In medical imaging,
CNNs have been effective in extracting complex features and patterns,
making them well- suited for scoliosis detection.
Literature review
Scoliosi Diagnostic Machine
s Method Learning
To diagnose and assess the degree Machine learning is a branch
Scoliosis is defined as a disease which of scoliosis progression of artificial intelligence (AI)
affects the spine with a Cobb angle of is a posturographic X-ray
and computer science which
at least 10° or more by the Scoliosis examination (stitching) that covers
focuses on the use of data and
Research Society (SRC). Scoliosis the entire spinal column. This examination
is traditionally performed in both algorithms to imitate the way
curvatures range from 5° to around
80°. anterior– posterior and lateral. This allows that humans learn, gradually
About 30% of people with scoliosis
not only the degree of curvature to be improving its accuracy. This
determined but also the degree of causes for machines to get to
will simply need to wear braces, while
rotation, the dynamics of the deformation
about 10% may require surgery. understand even better as their
process, and many others which makes it
way of thinking becomes more
easier to collect sufficient data from the
examination. similar to that of a human’s.
Machine Learning in CNN SVM
Medical Imaging Method Method
In medical imaging, machine learning

Convolutional Neural Networks Support Vector Machines
plays a crucial role in analyzing various (CNNs) have shown enormous (SVM) constitute a prominent
medical images. What they can do is potential recently in the field of machine learning technique
analyse and detect the abnormality in medical image processing. Unlike applied in scoliosis detection
the patterns. Because of this, machine most machine learning from backbone X-ray images.
learning was said to be able to perform techniques, CNN does not require This method proves to be
as well as most medical experts. any handcrafted coaching effective when analysing very
choices. With that fact stands, it complex patterns and
becomes more effective in structures.
choosing the suitable section of
a spine to be analysed.
Methodology
The entire research process involves meticulous data

acquisition, preprocessing, feature extraction, model training,
and an in-depth evaluation of model performance.
Data Acquistion involves bunch of X-ray pictures sourced from

Mendeley data. The dataset contains image of scoliosis and
normal backbone with varying degrees to make data have
more variety.
Data pre-processing will make sure all X-ray images have same size, adjusted
the brightness and fixed any abnormalities in the dataset. It make SVM and
CNN understand the images easier.
Feature Extraction will help models learn in which SVM focuses on spine
curvature characteristics and CNN learned by raw picture data.
SVM and CNN will be trained seperately. SVM is better at recognizing patterns
related to scoliosis while CNN learn complex patterns using layers with
hierarchical nature of neural networks.
Model Validation will use different dataset to handle new cases that have not
been encountered yet.
Performance Metrics evaluates measures such as accuracy, precision, recall and

F1 score to check how good the models were . ROC Curves are also used to
tell difference between scoliosis and normal at different levels.
Finally there is a comparative analysis to identify strength and weakness of SVM

and CNN in scoliosis detection to find scenario where one is better than the
other.
Results and Analysis (CNN)
Confusion Matrix Classification Report ROC Analysis

CNN Confusion Matrix shows that it Normal Class achieved ROC curves for CNN have
correctly identiified 11 Normal and 34 exceptional precision of 1.00 a compelling area under
Scoliosis cases in top-left and bottom- but a recall of 0.65 with a the curve (AUC-ROC)
right quadrant. balanced F1 score of 0.79. score of 0.9273.
However, 6 Normal cases are identified Scoliosis Class achieved It indicates a strong ability
as Scoliosis shown in top-right robust precision of 0.85 and of the CNN to distinguish
quadrant. a perfect recall of 1.00 with between normal and
Although, 0 Scoliosis cases are identified high F1 score of 0.92. scoliotic instances.
as Normal shown in bottom-left
quadrant.
Results and Analysis (SVM)
Confusion Matrix Classification Report ROC Analysis

SVM Confusion Matrix shows that it Normal Class achieved ROC curves for SVM have
correctly identiified 13 Normal and 31 substantial precision of 0.81 but a compelling area under
Scoliosis cases in top-left and bottom- a recall of 0.72 with a the curve (AUC-ROC)
right quadrant. balanced F1 score of 0.76. score of 0.9361.
However, 5 Normal cases are identified Scoliosis Class achieved It indicates a strong ability
as Scoliosis shown in top-right robust precision of 0.86 and of the SVM to distinguish
quadrant. a high recall of 0.91 with between normal and
Also, 3 Scoliosis cases are identified impressive F1 score of 0.89. scoliotic instances.
as Normal shown in bottom-left
quadrant.
Comparative Analysis
Convolutional Neural Support Vector

Networks(CNN) Machines (SVM)
Normal: Scoliosis: Normal Scoliosis:
Accuracy: Accuracy: : Accuracy: Accuracy:
1.00 0.85 0.81 0.86
Recall: 0.65 Recall: 1.00 Recall: 0.72 Recall: 0.91
F1-score: F1-score: F1-score: F1-score:
0.79 0.92 0.76 0.89
AUC-ROC score: 0.9273 AUC-ROC score: 0.9361
Has a balanced trade-off in detecting normal spines, but Can detect normal spines and scoliosis at a relatively
can accurately detect spines with scoliosis. average capability. However, having a higher AUC-ROC score
than CNN’s proves that SVM excels in superior
discriminatory power and effectiveness in clinical
applications.
Conclusions
The CNN model was able to get a precise and accurate result, as it excelled in precision
for the ”Normal” class and achieved a balanced trade-off between precision and recall.
Meanwhile, the SVM model demonstrated robust performance, particularly in precision
and discriminatory power, as indicated by the higher AUC-ROC score.
Both models show each of their pros and cons in providing their best analysis result,
but ultimately they are useful depending on the situation. As time passes, there might
be more improvements to them that allows them to be used for even more certain
scenarios.
Thank you

A Comparative Analysis of SVM and CNN For Scoliosis Detection in Backbone X-Ray Images

Uploaded by

Copyright:

Available Formats

A Comparative Analysis of SVM and CNN For Scoliosis Detection in Backbone X-Ray Images

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Comparative Analysis of SVM and CNN For Scoliosis Detection in Backbone X-Ray Images

Uploaded by

Copyright:

Available Formats

A Comparative Analysis of SVM

and CNN for Scoliosis Detection

AFFILIATION: BINA NUSANTARA

However, it is still undeniable that early detection still proves to be difficult.

Convolutional Neural Networks (CNN), a type of deep learning model, have

In medical imaging, machine learning

The entire research process involves meticulous data

Data Acquistion involves bunch of X-ray pictures sourced from

Performance Metrics evaluates measures such as accuracy, precision, recall and

Finally there is a comparative analysis to identify strength and weakness of SVM

Confusion Matrix Classification Report ROC Analysis

Confusion Matrix Classification Report ROC Analysis

Convolutional Neural Support Vector

You might also like