1. Introduction
The progressive and constant aging of the population causes a higher incidence of neurodegenerative diseases associated with age. Among these diseases, the Alzheimer’s has a prevalence of 5.05% in Europe as early as 2016 [
1]. Early palliative diagnosis and treatment continue to be the best alternatives for improving the patient’s quality of life and their environment. To make this diagnosis, various cognitive, psychological or clinical tests are used. Within the group of clinical tests, one of the most widely used is the analysis of brain images obtained by magnetic resonance imaging (MRI), as changes in brain morphology can be seen such as contraction of the hippocampus and cerebral cortex, or elongation of the ventricles [
2].
The objective of this work is the use of Deep Learning techniques to support an early diagnosis of Alzheimer’s disease through the analysis of conventional sagittal MRI images from two reference sets of data.
2. Materials and Methods
This section describes the datasets and computer models proposed in this work, each of which is described in its corresponding subsection.
2.1. Materials
This work makes the MRI images dataset ADNI [
3]. Both sets are collections of correctly labeled MRI images of 255 × 255 pixels, which are two of the most common in the literature.
2.2. Proposed Model
The proposed method uses the classical pipeline to solve a problem on any kind of signal, and in particular in this case images. This pipeline has a pre-established set of stages, being: pre-processing phase, feature extraction phase and a regression or classification phase based on the features extracted from the images.
The proposed model is based on the use of a pre-trained model which, in this particular case is ResNet [
4]. The idea behind this is to take advance of features extration phase of the model while the classification is droped or ajusted for a new problem. This squema known under the name of transfer learning has been used many times in the related literature. In the proposed model, the classifier has been replaced by a Support Vector Machine (SVM) [
5]. Thus, the speed of experimentation can be accelerated by adapting to this problem, successful models in other different problems.
3. Results
In this paper, the results for the ADNI dataset are presented. The results shown on
Table 1 are average of 50 repretitions of a Hold-Out training strategy. The table showns the test result comparison between a reference work and two developed developped models, one with ResNet as base and one with MobileNet [
6]. The dataset was splited in 80% for training and 20% for testing, while a 1% of the training data set was used for validation purposes. The advantages of the ResNet approach are noticeable being the best one in precision and recall which are our main objective.
4. Discussion
As a main conclusion the identification of Alzheimer’s disease in sagittal MRI images from ADNI dataset is accessible using Deep Learning techniques. These results are comparable to those proposed by the horizontal cuts in the literature. Despite the high imbalance of both data sets and the small OASIS set, the proposed model presents satisfactory results because of its simplicity.
Based on the authors’ experience in the field of Alzheimer’s disease, the sagittal plane also shows characteristic deformations of the disease. Traditionally, specialists use the horizontal plane. This opens up new ways for experimentation. New characteristics of Alzheimer’s disease can be found in them in other regions. These new features may make the diagnosis of Alzheimer’s a more precise task.
Author Contributions
Conceptualization, A.P.-C.; methodology, A.P.-C.; experiment design, C.M.; software, A.P.-C. and E.F.-B.; validation, A.P.-C.; formal analysis, A.P.-C.; investigation, A.P.-C.; resources, A.P.-C.; data curation, A.P.-C.; writing—original draft preparation, A.P.-C.; writing—review and editing, E.F.-B.; visualization, A.P.-C.; supervision, C.M. and E.F.-B.
Funding
This research received no external funding.
Acknowledgments
The authors would like to thank the support from NVidia corp., which granted the GPU used in this work. They also acknowledge the support from the CESGA, where many of the preliminary tests were run.
Conflicts of Interest
The authors declare no conflict of interest.
Abbreviations
The following abbreviations are used in this manuscript:
ANN | Artificial Neural Network |
MRI | Magnetic Resonance Imaging |
AUC | Area Under Curve ROC |
ROC | Receiver Operating Characteristic |
References
- Niu, H.; Álvarez-Álvarez, I.; Guillén-Grima, F.; Aguinaga-Ontoso, I. Prevalencia e incidencia de la enfermedad de Alzheimer en Europa: Metaanálisis. Neurología 2017, 32, 523–532. [Google Scholar] [CrossRef] [PubMed]
- Sarraf, S.; Tofighi, G. DeepAD: Alzheimer’s disease classification via deep convolutional neural networks using MRI and fMRI. BioRxiv 2016. p. 070441. [Google Scholar]
- ADNI|Alzheimer’s Disease Neuroimaging Initiative. Available online: http://adni.loni.usc.edu (accessed on 27 June 2019).
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
- Joachims, T. Text categorization with support vector machines: Learning with many relevant features. In European Conference on Machine Learning; Springer: Berlin/Heidelberg, Germany, 1998; pp. 137–142. [Google Scholar]
- Howard, A.G.; Zhu, M.; Chen, B.; Kalenichenko, D.; Wang, W.; Weyand, T.; Andreetto, M.; Adam, H. Mobilenets: Efficient Convolutional Neural Networks for Mobile Vision Applications. arXiv 2017, arXiv:1704.04861. [Google Scholar]
- Ding, Y.; Sohn, J.H.; Kawczynski, M.G.; Trivedi, H.; Harnish, R.; Jenkins, N.W.; Lituiev, D.; Copeland, T.P.; Aboian, M.S.; Mari Aparici, C.; et al. A Deep learning model to predict a diagnosis of alzheimer disease by using 18F-FDG PET of the brain. Radiology 2018, 290, 456–464. [Google Scholar] [CrossRef] [PubMed]
Table 1.
Best results for ADNI.
Table 1.
Best results for ADNI.
Model | Image | Accuracy | Precision | Recall | Specificity | f1-Score | AUC |
---|
Inception [7] | PET hor. | - | 63.66% | 64.67% | 79.00% | 64.00% | 76.00% |
ResNet | MRI sag. | 81.46%±1.9% | 82.48%±2.2% | 93.09%±1.9% | 55.19%±4.1% | 87.44%±1.5% | 74.14%±2.2% |
MobileNet | MRI sag. | 51.08%±19.1% | 37.56%±34.7% | 52.4%±49.7% | 47.73%±40.7% | 43.16%±40.7% | 50.01%±0.1% |
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).