Automatic Detection of Alzheimer’s Disease from Magnetic Resonance Images Using a Hybrid Deep Learning Approach

Öznur Özaltın; Sezgi Çobanbaş; Yasemin Sırakaya; Yuşa Güneş

Article Details

Automatic Detection of Alzheimer’s Disease from Magnetic Resonance Images Using a Hybrid Deep Learning Approach

Authors:

Öznur Özaltın | Sezgi Çobanbaş | Yasemin Sırakaya | Yuşa Güneş

Publisher:

Fırat University

Main Topics:

Derin Öğrenme

Journal:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.321-339

Document Type:

Article

Publication Date:

3/2025

Document Language:

Turkish

Document ID (Org No.):

2001334874

pISSN:

1308-9072

Keywords:

Abstract:

 Alzheimers disease is one of the most serious diseases of our age. Early diagnosis is very important in this almost incurable disease. The progress of this disease will be slowed down by early diagnosis. This study goals to identify Alzheimers disease based on deep learning from Magnetic Resonance MR images. Convolutional Neural Networks CNN, one of the deep learning algorithms, can automatically extract features from images. In this study, eight different CNN architectures were used to detect Alzheimers disease from MR images. NasNetMobile, the most successful of these architectures, was used as an automatic feature extractor, and 1056 features were extracted from its last layer. These features were classified through machine learning algorithms: Deep Neural Network DNN, Support Vector Machine SVM, Decision Tree, K-Nearest Neighbor, and Artificial Neural Network ANN. Although high accuracy was achieved at this stage of the study, the minimum Redundancy Maximum Relevance mRMR approach was used for selecting features and robust results. Thus, 1056 features were reduced to 250. As a result, the classification performance of machine learning algorithms improved. The NasNetMobile-mRMR-DNN hybrid algorithm was the most successful in the detection of Alzheimers disease with an accuracy rate of 90.68.

 Alzheimers disease is one of the most serious diseases of our age. Early diagnosis is very important in this almost incurable disease. The progress of this disease will be slowed down by early diagnosis. This study goals to identify Alzheimers disease based on deep learning from Magnetic Resonance MR images. Convolutional Neural Networks CNN, one of the deep learning algorithms, can automatically extract features from images. In this study, eight different CNN architectures were used to detect Alzheimers disease from MR images. NasNetMobile, the most successful of these architectures, was used as an automatic feature extractor, and 1056 features were extracted from its last layer. These features were classified through machine learning algorithms: Deep Neural Network DNN, Support Vector Machine SVM, Decision Tree, K-Nearest Neighbor, and Artificial Neural Network ANN. Although high accuracy was achieved at this stage of the study, the minimum Redundancy Maximum Relevance mRMR approach was used for selecting features and robust results. Thus, 1056 features were reduced to 250. As a result, the classification performance of machine learning algorithms improved. The NasNetMobile-mRMR-DNN hybrid algorithm was the most successful in the detection of Alzheimers disease with an accuracy rate of 90.68.

 Alzheimers disease is one of the most serious diseases of our age. Early diagnosis is very important in this almost incurable disease. The progress of this disease will be slowed down by early diagnosis. This study goals to identify Alzheimers disease based on deep learning from Magnetic Resonance MR images. Convolutional Neural Networks CNN, one of the deep learning algorithms, can automatically extract features from images. In this study, eight different CNN architectures were used to detect Alzheimers disease from MR images. NasNetMobile, the most successful of these architectures, was used as an automatic feature extractor, and 1056 features were extracted from its last layer. These features were classified through machine learning algorithms: Deep Neural Network DNN, Support Vector Machine SVM, Decision Tree, K-Nearest Neighbor, and Artificial Neural Network ANN. Although high accuracy was achieved at this stage of the study, the minimum Redundancy Maximum Relevance mRMR approach was used for selecting features and robust results. Thus, 1056 features were reduced to 250. As a result, the classification performance of machine learning algorithms improved. The NasNetMobile-mRMR-DNN hybrid algorithm was the most successful in the detection of Alzheimers disease with an accuracy rate of 90.68.

 Alzheimer hastalığı, çağın en ciddi hastalıkları arasında yer almaktadır. Tedavisinin neredeyse bulunmadığı bu hastalıkta erken teşhis çok önemlidir. Erken teşhis edilmesi durumunda bu hastalığın ilerlemesi yavaşlayacaktır. Bu çalışmada, Alzheimer hastalığının derin öğrenme aracılığı ile Manyetik Rezonans MR görüntüler üzerinden tespit edilmesi amaçlanmıştır. Derin öğrenme algoritmalarından olan Evrişimli Sinir Ağları ESA, görüntülerden otomatik özellik çıkarabilme yeteneğine sahiptir. Bu çalışmada, Alzheimer hastalığını MR görüntülerinden tespit etmede sekiz farklı ESA mimarisi kullanılmıştır. Bu mimarilerden en başarılı test sonucu veren NasNetMobile, otomatik özellik çıkarıcı olarak kullanılmış ve son katmanından 1056 özellik elde edilmiştir. Bu özellikler makine öğrenme algoritmalarından Derin Sinir Ağı DSA, Destek Vektör Makinesi DVM, Karar Ağacı, K-En Yakın Komşu, Yapay Sinir Ağı YSA ile sınıflandırılmıştır. Çalışmanın bu aşamasında yüksek doğruluk oranı elde edilse de sağlam sonuçlar için özellik seçim yöntemlerinden minimum Artıklık Maksimum İlişki mRMR yaklaşımından yararlanılmıştır. Böylece 1056 özellik 250ye indirgenmiştir. Sonuç olarak, makine öğrenme algoritmalarının sınıflandırma performansı artmıştır. Alzheimer hastalığının tespitinde 90,68lik doğruluk oranı ile en başarılı sonuç veren NasNetMobile-mRMR-DSA hibrit algoritması olmuştur.

Article Details

Automatic Detection of Alzheimer’s Disease from Magnetic Resonance Images Using a Hybrid Deep Learning Approach

Authors:

Öznur Özaltın | Sezgi Çobanbaş | Yasemin Sırakaya | Yuşa Güneş

Publisher:

Fırat University

Main Topics:

Derin Öğrenme

Journal:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.321-339

Document Type:

Article

Publication Date:

3/2025

Document Language:

Turkish

Document ID (Org No.):

2001334874

pISSN:

1308-9072

Alzheimer’s Disease

Deep Learning

Machine Learning

Feature Selection

Classification.Alzheimer Hastalığı

Derin Öğrenme

Makine Öğrenmesi

Özellik Seçimi

Sınıflandırma

Issues

Article Details

Authors: Öznur Özaltın | Sezgi Çobanbaş | Yasemin Sırakaya | Yuşa Güneş

Publisher: Fırat University

Main Topics: Derin Öğrenme

Journal: Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.321-339

Document Type: Article

Publication Date: 3/2025

Document Language: Turkish

Document ID (Org No.): 2001334874

pISSN: 1308-9072

Issues

Authors:

Öznur Özaltın | Sezgi Çobanbaş | Yasemin Sırakaya | Yuşa Güneş

Publisher:

Fırat University

Main Topics:

Derin Öğrenme

Journal:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.321-339

Document Type:

Article

Publication Date:

3/2025

Document Language:

Turkish

Document ID (Org No.):

2001334874

pISSN:

1308-9072