Brain Matters Vol. 8 No. 1


Written by Rayyan Iqbal

The Potential of Hybrid
Models in Alzheimer's
Diagnosis: Combining Neural
Networks and SVMs for
Enhanced Accuracy 

Introduction
Alzheimer’s Disease (AD) affects over 55 million people
worldwide, posing a major challenge for healthcare systems
as populations continue to age. As a progressive
neurodegenerative disorder, AD leads to severe cognitive
decline, memory loss, and a profound reduction in quality of
life, ultimately resulting in death. The mortality rate for AD
approaches 100%, with patients typically living only 3 to 11
years after diagnosis, underscoring the critical importance
of early detection (Alzheimer’s Stages: How the Disease
Progresses, n.d.). Despite extensive research, early diagnosis
of AD remains difficult.

The need for early detection is clear when observing
cognitive tests like clock-drawing exercises. The differences
between healthy individuals and those in late-stage AD are
striking; however, distinguishing early-stage AD from
normal aging is much harder, as seen in the clock drawings,  
in which there isn’t as profound of a difference in the quality
of the clock drawn between normal and early AD individuals
(Mattson 2014). Additionally, brain degeneration progresses
significantly as AD advances, as shown in the brain
degeneration image (Maha 2023). While late-stage AD
presents notable structural changes, the brain deterioration
in early stages is minimal, making it challenging to identify
AD before symptoms become severe.

Traditional diagnostic methods—including clinical
assessments, cognitive tests, and neuroimaging—often
struggle to detect AD at these early stages, delaying the
possibility of effective interventions. Recent advancements
in machine learning offer new hope. Neural networks (NNs),
particularly convolutional neural networks (CNNs), excel at
extracting complex patterns from large datasets, such as
MRI or PET scans, allowing for a more nuanced analysis of
brain images (Taherdoost 2023). Support vector machines
(SVMs), powerful classification tools, complement this
process by distinguishing between healthy individuals and
those with various degrees of cognitive impairment
(Ahmadi et al., 2024).

While each model has shown promise independently, hybrid
models that combine the strengths of CNNs and SVMs could
further improve diagnostic accuracy. Hybrid models
leverage CNNs’ ability to identify intricate patterns and
SVMs’ adeptness at classification, potentially making early
detection achievable (Li 2023). However, research in this
area is still limited, with few large-scale studies validating
these methods. Before these hybrid models can be
integrated into clinical practice, more robust data and
validation are necessary to ensure their reliability. While the
potential of these hybrid models is promising, caution is
urged in their application until further research solidifies
their effectiveness.

18



Neural Networks: Identifying Early-Stage
Alzheimer’s in Neuroimaging
Neural networks, especially convolutional neural networks
(CNNs), have proven highly effective at analyzing
neuroimaging data like MRI or PET scans, which is essential
for diagnosing Alzheimer’s Disease in its early stages. Neural
networks are inspired by the human brain’s structure,
consisting of interconnected layers of “neurons” that learn
from data by recognizing patterns and relationships. When
given input data (such as numbers, images, or text), the
network adjusts the connections between these neurons to
improve its understanding. This allows neural networks to
perform tasks such as predicting outcomes, classifying
objects, and identifying trends (Taherdoost 2023). 

CNNs are a specialized type of neural network designed
specifically for image data. They use a process called
“convolution,” where small filters scan sections of an image
to identify important features like edges, shapes, or
textures. Rather than analyzing the entire image at once,
CNNs focus on small regions at a time, making them
exceptionally good at interpreting visual information. In
simple terms, while a standard neural network might act as
a general “decision-maker,” CNNs operate more like a team
of “spotters,” with each focusing on a different part of the
image to capture crucial details (e.g., one “spotter” might
identify the nose, another the eyes). This approach allows
CNNs to excel at tasks such as recognizing objects in photos
or detecting abnormalities in medical images. 

CNNs can achieve remarkable accuracy in identifying early
signs of Alzheimer’s. In one study using a dataset of healthy
individuals, those with early Alzheimer’s, and those with
late-stage Alzheimer’s, a CNN model called ResNet-18
achieved an impressive 96.85% accuracy in distinguishing
between different stages of the disease based on MRI and
PET scans (Odusami et al., 2021). This performance far
exceeds conventional diagnostic methods, such as physician
evaluations of PET scans, which typically achieve about 85%
accuracy, according to the Texas Department of State Health
Services (Texas DSHS, 2021). 

However, CNNs have limitations. One major drawback is
their “black-box” nature, meaning they generate results
without revealing how they reached those conclusions. This
lack of transparency can be problematic in clinical settings,
where doctors need to understand the reasoning behind a
diagnosis to justify treatment decisions. Without insight
into how CNNs make their determinations, it becomes
challenging for physicians to fully trust and adopt these
models in practice, despite their accuracy (Patil et al., 2022). 

SVMs: Precise Classification but Limited in
Complex Data 
While CNNs are excellent for identifying features in images,
support vector machines (SVMs) are highly effective for
classifying data once those features are identified. The
image above illustrates how an SVM functions: two sets of
data points, shown as green and red dots, represent two
different classes the SVM aims to separate. The axes, labeled
x1 and x2, represent a two-dimensional feature space where
the data points are plotted. The blue line running diagonally
across the plot is the decision boundary, which separates the
two classes. This boundary isn’t just any line—it is chosen to
maximize the distance between itself and the nearest data
points from each class, a concept known as the “margin.”
One of the points closest to the boundary, called a “support
vector” (represented by the purple dot), plays a crucial role
in defining this margin (Tan 2020). 

The Potential of Hybrid Models in Alzheimer's Diagnosis: Combining Neural Networks and SVMs for Enhanced Accuracy

Figure 1. Clock drawings showing cognitive decline:
Normal (accurate), Early AD (slightly distorted), Late AD

(severely disorganized). (Linus Health) 

Figure 2. Brain atrophy progression: Healthy, Early
Alzheimer's (mild shrinkage), Severe Alzheimer's

(significant shrinkage). (First Choice Neurology 2019). 

19



SVMs have demonstrated reliable performance in classifying
patients by analyzing MRI data, with their highest accuracy
reaching 89%. They excel in identifying key features like
gray matter volume and brain asymmetry to distinguish
between control groups, early mild cognitive impairment
(eMCI), late mild cognitive impairment (lMCI), and
Alzheimer’s disease stages. In one approach, researchers
used hippocampal volume—a primary indicator of
Alzheimer’s—as the main feature, which contributed to the
high accuracy achieved in their study (Ahmadi et al., 2024). 

 Despite their strengths, SVMs face challenges, particularly
with high-dimensional data, where each data point has
numerous characteristics or “dimensions.” In such cases, it
becomes harder for SVMs to identify clear patterns or find a
meaningful decision boundary. Without careful feature
selection and model tuning, SVMs may struggle to classify
complex data accurately, leading to overfitting. Overfitting
occurs when a model performs well on training data but
poorly on new, unseen data, reducing its real-world
effectiveness (Ameen et al., 2024).

Hybrid Models: Combining Neural Networks
and SVMs for Enhanced Accuracy

In recent years, hybrid models that combine convolutional
neural networks (CNNs) and support vector machines
(SVMs) have gained attention in Alzheimer’s disease
diagnosis, offering a more comprehensive approach to early
detection and classification (Li 2023). By integrating CNNs
and SVMs, these models provide a robust framework to
handle the complexity of high-dimensional neuroimaging
data, which is essential for identifying subtle early-stage
markers of Alzheimer’s. 

Hybrid models leverage the unique strengths of CNNs and
SVMs. CNNs are particularly skilled at extracting features

Brain Matters Vol. 8 2025

Figure 3. Support Vector Machine (SVM) diagram showing
a decision boundary separating two classes. (Tan 2020). 

from complex neuroimaging data, such as magnetic
resonance imaging (MRI) and positron emission tomography
(PET) scans. These models can detect early
neurodegenerative changes, including hippocampal
atrophy, cortical thinning, and ventricular enlargement—
key biomarkers for diagnosing Alzheimer’s at its early
stages (Oostven 2021). However, while CNNs excel at
identifying these features, they often lack the precision
needed for effective classification. This is where SVMs
become valuable, as they create precise boundaries that
distinguish different cognitive states, such as healthy
controls, patients with mild cognitive impairment (MCI),
and those with Alzheimer’s disease (Ahmadi et al., 2024). 

 A study by Al Subaie et al. demonstrated the effectiveness
of a CNN-SVM hybrid model applied to neuroimaging data,
showing the model’s ability to distinguish between MCI
patients likely to progress to Alzheimer’s and those who are
not. This research highlighted the advantage of combining
CNNs’ feature extraction capabilities with SVMs’
classification strength. The hybrid model achieved a higher
accuracy of 98.20%, compared to 91.70% for models relying
solely on CNNs or SVMs. The model was particularly
successful in detecting subtle brain changes indicative of
MCI conversion, which is essential for timely intervention
(Al Subaie et al., 2024). 

 Basheera et al. explored hybrid deep learning models using
multimodal neuroimaging data, combining MRI and PET
scans (Basheera et al., 2019). In this approach, the CNN
component analyzed the raw imaging data to extract
patterns of brain degeneration, which the SVM then
classified into diagnostic categories like non-demented,
MCI, and AD. This hybrid method demonstrated superior
diagnostic accuracy, especially in differentiating between
early MCI and advanced cognitive decline stages. The results
showed that the CNN-SVM model outperformed traditional
diagnostic methods and standalone machine-learning
models in terms of accuracy, sensitivity, and specificity,
achieving 90.47% accuracy, 86.66% recall, and 92.59%
precision (Basheera et al., 2019).

Both studies emphasize the potential of hybrid models to
bridge the gap between feature extraction and classification,
especially when dealing with high-dimensional
neuroimaging data. By allowing CNNs to identify complex
biomarkers and enabling SVMs to classify these features
accurately, hybrid models offer a more nuanced approach to
diagnosing Alzheimer’s disease. This combination captures
subtle brain patterns that conventional methods might
miss, which is crucial for early detection. With earlier
intervention, hybrid models could lead to more accurate
diagnoses and better treatment options, potentially
improving patient outcomes. 

20



Considerations and Future Directions:
Evaluating the Potential of Hybrid Models for
Clinical Adoption 
While hybrid models combining neural networks (NNs) and
support vector machines (SVMs) show significant potential
in improving the accuracy of Alzheimer’s diagnosis, several
challenges currently limit their adoption in clinical practice.

One major challenge is the complexity and quality of data
required for these models to perform effectively. CNNs, a
critical component of hybrid models, rely heavily on
neuroimaging data, which is often difficult and costly to
obtain at the scale needed for robust machine learning.
Studies, including those focusing on multimodal
approaches, have highlighted that even well-curated
datasets like the Alzheimer’s Disease Neuroimaging
Initiative (ADNI) are often insufficiently large or diverse to
ensure the generalizability of CNNs across various
populations. Additionally, collecting such data is
complicated by privacy concerns, which can further limit
the availability of comprehensive datasets (Ismail et al.,
2022). 

Another significant challenge is the interpretability of
hybrid models. Neural networks, especially CNNs, often
function as “black boxes,” producing diagnostic results
without revealing how they arrived at those conclusions.
This lack of transparency makes it difficult for healthcare
professionals to understand and trust the model’s rationale,
particularly when making critical patient care decisions.
While efforts like Explainable AI have aimed to make
machine learning models more interpretable, the need for
transparent decision-making remains a major barrier to
adoption in clinical settings (Prijs et al., 2022). 

In addition, hybrid models are prone to overfitting—a
problem where the model performs well on training data
but poorly on new, unseen data. This is especially relevant
for models that incorporate both SVMs and CNNs, as SVMs
can struggle with high-dimensional data without careful
feature selection, and CNNs require extensive
computational resources and large datasets to avoid
overfitting. If not properly managed, overfitting can
undermine the model’s real-world effectiveness and limit its
reliability in clinical settings (Ying 2019). 

Overall, while hybrid models hold great promise, they are
not yet ready for widespread clinical adoption. More
research is needed to validate their performance across
larger, more diverse datasets and to address issues related
to interpretability and scalability. Addressing these
challenges will be essential for realizing the full potential of
hybrid models in Alzheimer’s diagnosis and ensuring they
can be safely and effectively integrated into everyday
clinical practice (Wang et al., 2024).

Conclusion
The integration of neural networks and support vector
machines represents a promising approach to improving
early Alzheimer’s diagnosis. Neural networks, particularly
convolutional neural networks (CNNs), excel at identifying
subtle patterns in neuroimaging data, while SVMs provide
precise classification capabilities. By combining these
techniques in hybrid models, diagnostic accuracy can be
significantly enhanced, potentially enabling earlier
intervention and improved patient outcomes (Oostven et.
al., 2021). 

However, despite these advancements, the field remains
under-researched. The studies conducted so far, while
encouraging, are not yet sufficient to support the
widespread adoption of hybrid models in clinical practice.
More research is necessary to validate these models on
larger and more diverse datasets and to address their
limitations, particularly in terms of neural network
interpretability and the extensive feature selection required
for SVMs. 

As machine learning continues to evolve, hybrid models
combining neural networks and SVMs may ultimately
transform Alzheimer’s diagnosis, enabling more accurate
and timely detection. For now, however, these methods
should be approached with caution, as further rigorous and
repetitive studies are needed to confirm their reliability in
clinical settings. With continued research, these models
could one day offer a breakthrough in diagnosing and
managing Alzheimer’s disease. 

References 
1. Ahmadi, M., Javaheri, D., Khajavi, M., Danesh, K., & Hur, J.
(2024). A deeply supervised adaptable neural network for
diagnosis and classification of Alzheimer’s severity using
multitask feature extraction. PLoS ONE, 19(3), e0297996.
https://doi.org/10.1371/journal.pone.0297996 
2. Ameen, T. B., Kashif, S. N., Abbas, S. M. I., Babar, K., Ali, S.
M. S., & Raheem, A. (2024). Unraveling Alzheimer’s: the
promise of aducanumab, lecanemab, and donanemab. The
Egyptian Journal of Neurology Psychiatry and
Neurosurgery, 60(1). https://doi.org/10.1186/s41983-024-
00845-5 
3. Alsubaie, M. G., Luo, S., & Shaukat, K. (2024). Alzheimer’s
Disease Detection Using Deep Learning on Neuroimaging: A
Systematic review. Machine Learning and Knowledge
Extraction, 6(1), 464–505.
https://doi.org/10.3390/make6010024
4. Basheera, S., & Ram, M. S. S. (2019). Convolution neural
network–based Alzheimer’s disease classification using
hybrid enhanced independent component analysis based
segmented gray matter of T2 weighted magnetic resonance
imaging with clinical valuation. Alzheimer’s & Dementia:
Translational Research & Clinical Interventions, 5(1), 974–
986. https://doi.org/10.1016/j.trci.2019.10.001

21

The Potential of Hybrid Models in Alzheimer's Diagnosis: Combining Neural Networks and SVMs for Enhanced Accuracy



5. Ismail, W. N., PP, F. R., & Ali, M. A. S. (2022). MULTFORAD:
Multimodal MRI neuroimaging for Alzheimer’s disease
detection based on a 3D convolution model. Electronics,
11(23), 3893. https://doi.org/10.3390/electronics11233893 
6. Li, M., Jiang, Y., Zhang, Y., & Zhu, H. (2023). Medical image
analysis using deep learning algorithms. Frontiers in Public
Health, 11, 1273253.
https://doi.org/10.3389/fpubh.2023.1273253 
7. Odusami, M., Maskeliūnas, R., Damaševičius, R., &
Krilavičius, T. (2021). Analysis of Features of Alzheimer’s
Disease: Detection of Early Stage from Functional Brain
Changes in Magnetic Resonance Images Using a Finetuned
ResNet18 Network. Diagnostics, 11(6), 1071.
https://doi.org/10.3390/diagnostics11061071 
8. Patil, V., Madgi, M., & Kiran, A. (2022). Early prediction of
Alzheimer’s disease using convolutional neural network: a
review. The Egyptian Journal of Neurology Psychiatry and
Neurosurgery, 58(1). https://doi.org/10.1186/s41983-022-
00571-w 
9. Taherdoost, H. (2023). Deep Learning and Neural
Networks: Decision-Making Implications. Symmetry, 15(9),
1723. https://doi.org/10.3390/sym15091723 
10. Van Oostveen, W. M., & De Lange, E. C. M. (2021). Imaging
Techniques in Alzheimer’s Disease: A review of applications
in early diagnosis and longitudinal monitoring.
International Journal of Molecular Sciences, 22(4), 2110.
https://doi.org/10.3390/ijms22042110 
11. Wang, Y., Gao, R., Wei, T., Johnston, L., Yuan, X., Zhang, Y.,
& Yu, Z. (2024). Predicting long-term progression of
Alzheimer’s disease using a multimodal deep learning model
incorporating interaction effects. Journal of Translational
Medicine, 22(1). https://doi.org/10.1186/s12967-024-05025-w 
12. Ying, X. (2019). An Overview of Overfitting and its
Solutions. Journal of Physics: Conference Series, 1168,
022022. https://doi.org/10.1088/1742-6596/1168/2/022022

Brain Matters Vol. 8 2025

Images: 
1. First Choice Neurology. (2019). June is Alzheimer’s and
Brain Awareness Month. First Choice Neurology. Retrieved
February 18, 2025, from 
https://www.fcneurology.net/june-is-alzheimers-and-
brain-awareness-month/ 
2. Linus Health. (n.d.). History of the Clock Drawing Test and
the Linus Health Platform. Linus Health. Retrieved February
18, 2025, from https://linushealth.com/history-of-clock-
drawing-test-and-dctclock 
3. Tan, A. (2020). Support Vector Machine. Anthony’s Blogs.
Retrieved February 18, 2025, from https://anthony-
tan.com/Discriminant-Functions-and-Decision-Boundary/

22

About the Author
Rayyan Iqbal is a sophomore at the University of Illinois, majoring in
Chemistry. He is currently conducting research in the Physical Activity
and Neurocognitive Health Lab, where he studies the impact of physical
behaviors—such as physical activity and sedentary time—on brain
health. Beyond his research, Rayyan is actively involved in REACT, an
outreach program that brings science to life for young students in the
Champaign-Urbana area.



23