CycleGAN-based Data Augmentation to Improve Generalizability Alzheimer’s Diagnosis using Deep Learning

CycleGAN-based Data Augmentation to Improve Generalizability Alzheimer’s Diagnosis using Deep Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article CycleGAN-based Data Augmentation to Improve Generalizability Alzheimer’s Diagnosis using Deep Learning Satish Kumar, Tasleem Arif This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4141650/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Alzheimer's disease is a degenerative condition that progressively damages brain neurons, ultimately leading to dementia and death. Despite the limited number of available samples, effective diagnostic methods are crucial to diagnose Alzheimer's disease. Typically, a combination of laboratory and neuro-psychological testing is employed for diagnosis. The decrease in brain mass linked to Alzheimer's disease can be identified by MRI scans, which makes it a suitable problem for deep learning and computer vision. A precise and effective deep learning model would provide physicians with valuable support for their diagnoses. However, medical data is often challenging to obtain, and deep learning requires considerable data. To address this issue, generative adversarial networks can be useful. In this study, we proposed a CycleGAN to generate relevant synthetic images of intestinal parasites to solve the data scarcity challenge. To classify Alzheimer's disease using MRI scans, we developed convolutional neural networks based on the Google Inceptionv3 CNN architecture for this study. We attained an impressive F-1 score of 89%. Furthermore, we demonstrated the effectiveness of GANs in enhancing classification accuracy when used for data augmentation by creating samples with CycleGAN, achieving a remarkable F-1 score of 95%. CycleGAN Data Augmentation MRI harmonization deep learning Alzheimer’s disease InceptionV3 Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1 Introduction Alzheimer's disease (AD) is a neurodegenerative condition primarily affecting the elderly population. According to Kimura et al. [ 1 ], it is characterized by a decline in memory, cognitive function, and behavioral function that significantly affects patients' daily activities. Studies show that the prevalence of AD is expected to triple globally by 2050, affecting over 100 million people [ 2 ]. According to [ 3 ][ 4 ], mild cognitive impairment (MCI) is a cognitive condition that is in between the early stages of Alzheimer's disease and normal ageing. Two types of MCI are described: progressive MCI (pMCI) and stable MCI (sMCI). Patients with sMCI remain in a constant state and may even recover in some situations, while those with pMCI progress to a diagnosis of AD [ 5 ][ 6 ]. Currently, there are no clinically useful drugs for treating AD. Therefore, the major focus of treatment has shifted towards early-stage AD diagnosis[ 7 ]. Identifying high-risk patients and implementing tailored treatment measures to slow down the disease's progression requires differentiating between those experiencing normal cognitive decline and those with AD, sMCI, or pMCI (various stages of AD)[ 8 ][ 9 ]. Moreover, the application of image processing techniques relevant to Alzheimer's disease, such as image segmentation and denoising, can aid in the early identification of the disease. In recent years, the work of [ 10 ] serves as an example of how the field of psychoradiological research has produced useful clinical data in favour of the identification of diagnostic and therapeutic neuroimaging markers for people with psychiatric disorders. Artificial intelligence and computational psychoradiology have become more common in the diagnosis of Alzheimer's disease (AD). Still, a larger number of well-processed, high-quality images are required in order to accurately diagnose AD. Sorin et al. [ 11 ], proposed the deep learning techniques used for image processing that to have some limitations. This show that the development of an efficient deep learning method that can process images is needed. A group under the direction of Goodfellow [ 12 ]unveiled an intriguing idea, the Generative Adversarial Network (GAN). In essence, GANs are deep learning models that are primarily intended to be used with images. The fundamental concept of GANs is to establish a kind of game in which the discriminator, on the one hand, tries to discern between actual and produced images, and the generator, on the other, attempts to create images. Sorin et al. [ 11 ] highlighted that the interaction between the discriminator and generator is fundamental to the functioning of GANs. Generative Adversarial Networks (GANs) have the potential to significantly enhance data augmentation, especially in the medical field where there is a scarcity of available data. Unlike conventional data augmentation techniques that modify existing images, GANs can generate new, similar images from a relatively small dataset. In our research, we are exploring the possibilities of using a convolutional neural network model for classifying Alzheimer's disease. Additionally, we are assessing the feasibility of enhancing our dataset using GANs, specifically the CycleGAN design. 2 Literature Review Brain MRI scans have been extensively utilized in previous machine learning studies to differentiate between elderly individuals suffering from Alzheimer's disease and those without cognitive problems who are of the same age. These studies often incorporate other imaging techniques or genetic data. Brain MRI scans have also been employed to distinguish between Alzheimer's patients and those suffering from other diseases or different types of dementia. Unsupervised clustering has been used to categorize patients with various forms of dementia [ 2 ]. This approach has been proven useful in predicting cognitive decline, such as the transition from mild cognitive impairment to Alzheimer's within a specific timeframe[ 13 ]. In the field of deep learning models, researchers Huang, Yechong, Xu, Jiahang, et al. have introduced a multi-task 3D CNN based on Convolutional Neural Networks (CNNs) [ 14 ]. This unique CNN not only utilizes an attention mechanism to classify Alzheimer's disease but it also estimates an individual's age using an MRI scan. This task of estimating age is often referred to as the brain age problem [ 2 ]. Additionally, variations on this fundamental CNN design exist. For instance, hybrid Recurrent Neural Networks (RNNs) can be used to collect data from several 2D MRI slices[ 15 ]. The researchers have also explored methods like pre-training and data augmentation to enhance the model's performance. A recent study has utilized a state-of-the-art deep convolutional neural network (CNN) architecture, InceptionResNet-V2, to analyze a vast number of MRI scans, totaling 85,721[ 16 ]. The researchers employed the pre-trained network to classify Alzheimer's disease, following a sex classification task. The results were remarkable, with accuracy rates exceeding 92% across three distinct, publicly available datasets. Furthermore, generative adversarial networks (GANs) offer significant possibilities for data augmentation and model architecture, as depicted in Fig. 1 . Rather than merely modifying existing data in the training set, GANs can generate synthetic data by extrapolating beyond it, effectively oversampling. A novel network, CycleGAN, developed by Zhu et al. [ 17 ], can perform image-to-image translation even when the images do not directly match up. This system enables two-way translation between the source domain X and the target domain Y, with two generator networks, G1 and G2, and their corresponding partner networks, D1 and D2, participating in an adversarial training process. While determining if a picture is synthetic or real, the discriminator (D) functions similarly to a decision maker. It contrasts a random authentic image from the target domain with the artificial image produced by the generator (G). It is G's responsibility to enhance the synthetic image's quality in order to trick the discriminator. After receiving an image from the source domain (x in X), the generator (G) creates a synthetic image (y = G(x)). After that, the discriminator (D) compares this artificial output to a randomly selected picture from the target domain (y in Y), as show in Fig. 2 . The generator's architecture consists of three components: a decoder, a transformer (usually a residual network), and an encoder. The discriminator uses a PatchGAN model to categorise photos as artificial or real. The learning rate was initially set to 0.002 for the first 50 epochs of training, then it was progressively lowered to zero after that. 3 Methodology The visual representation in Fig. 3 outlines the workflow of the model, which comprises of three primary phases: data collection, preprocessing utilizing CycleGAN-based augmentation, and classification via the Google InceptionV3 CNN model. 3.1 Data collection and pre-processing The experiment is based on information provided by the Institute for Information and Communications Technology Promotion (IITP) database. The primary objective of the Alzheimer's Disease Neuroimaging Initiative (ADNI) is to determine whether mild cognitive impairment (MCI) and early Alzheimer's disease (AD) can be tracked over time through a combination of clinical and neuropsychological assessments, positron emission tomography (PET), serial magnetic resonance imaging (MRI), and other biological markers. The study employed the ADNI1 standardized MRI dataset, which was segregated into three severity-based categories: AD, MCI, and normal cognition (NC). [ 18 ]. The primary aim of the study was to train a neural network to differentiate between AD and NC. To achieve this, the data was converted from NIFTI files to three-dimensional NumPy arrays using nibabel. Additionally, a CSV file was extracted from the ADNI database, which included scan data, patient specifics, and the actual diagnosis labels. The production of training samples was accomplished by slicing the three-dimensional NIFTI picture data and extracting three slices from the original image: one from the axial perspective, one from the coronal view, and one from the sagittal view. For uniformity, these slices were enlarged to 224 × 224 from the center of each axis length. Furthermore, to separate the brain tissue from the MRI pictures and improve sample consistency, the study employed a process known as "skull stripping," which involved using the deep brain library's Extractor function. In order to further refine the consistency of the data, RAS and ISO transforms and histogram normalization were applied using the TorchIO library[ 19 ]. These transformations changed the orientation of the MRI picture. 3.2 CycleGAN based data augmentation A CycleGAN model was employed in the study, comprising of two generators as depicted in Fig. 4 . One generator converted samples from normal cognition (NC) into Alzheimer's disease (AD) samples while the other generator did the opposite. During training, the first generator processed the real NC image to generate a fake AD image, which was then compared to a real AD image by a discriminator to calculate the GAN loss. The same training procedure was repeated with real AD images to yield another GAN loss. To build the original visuals, the fake images were passed through the second generator. A cycle consistency loss was then calculated by comparing the original NC image with its reconstruction, and repeating the process for the AD images to ensure the consistency of these reconstructions. The cycle consistency loss and GAN losses were combined to form the overall loss function, which aimed to maximize discriminators and minimize generators. The generator utilized a variety of techniques such as up-sampling, residual blocks, and down-sampling, in conjunction with the Google Inceptionv3 CNN architecture. The dataset was randomly paired after being divided into labels. Three distinct CycleGAN models were trained, each targeting a different MRI slice. These models were trained for 100 epochs with a batch size of 1, using the Adam optimizer. Subsequently, the trained model generated enough samples to produce a balanced dataset. For each NC sample, an AD version was created, and vice versa. Ultimately, 480 NC and 710 AD samples of each orientation were produced, resulting in a total of 1190 pictures for each class, as shown in table 1. Table 1. Dataset collection after CycleGAN Dataset Samples CycleGAN Generated Total Normal samples 710 480 1190 Alzheimer’s Samples 480 710 1190 Total 1190 1190 2380 3.3 Deep CNN Classifier In our study, we utilized the Google Inceptionv3 Convolutional Neural Network (CNN) architecture as a pre-trained model. Although the original architecture was designed for RGB images with three channels, our MRI scans are single-channel grayscale images. To adapt to the network, we replicated the same data three times along one dimension, converting them into three channels. Additionally, we modified the last layer to enable the network to function as a binary classifier. We employed a modified CNN with several inputs to effectively handle the volumetric data. The outputs from each of the three distinct Google Inceptionv3 CNN in this model are pooled and directed through fully connected layers to produce the diagnosis group at the end. The neural network underwent training for 100 epochs with a batch size of 16 using the Adam optimizer at a learning rate of .0002. We partitioned our data into three sets: 70% of the data was utilized for training, 20% for validation, and 10% for testing purposes. 3.4 Model Evaluation: In Table 2 , we present a compilation of various performance measures employed in this research. Table 2 Performance metrics used in deep learning studies Metric Formula \(\text{Accuracy}\) \(\frac{\text{TP+TN}}{\text{TP + TN + FP + FN }}\) \(\text{Recall}\) \(\frac{\text{TP}}{\text{TP + FN }}\) \(\text{Precision}\) \(\frac{\text{TP}}{\text{TP + FP }}\) \(\text{F1-Score}\) \(\text{2*}\frac{\text{Precion*Recall}}{\text{Precion+Recall}}\) 4 Results and Discussion The data augmentation process using CycleGAN for both NC and AD samples resulted in an additional 480 and 710 samples for each orientation, respectively. These results are presented in Table 3 , indicating that the total number of images for each class increased to 1190. We observed a significant improvement in the performance of the CNN model by employing GAN augmentation.[ 22 ]. The F1 score of the Google Inceptionv3 model increased by 9.4% from 0.900 to 0.942, indicating that the addition of CycleGAN greatly enhances the CNN's classification performance. Table 3 Comparison of CNN based models CycleGAN augmentation Models Accuracy Precision Recall F1 Score InspectionV3 0.824 0.748 0.852 0.811 InspectionV3 + GAN 0.918 0.880 0.921 0.900 InspectionV3 + CycleGAN 0.922 0.934 0.951 0.942 The synthesized images contained significant elements that are likely responsible for the beneficial effect. Moreover, the larger dataset and improved class balance in the CycleGAN-augmented data likely contributed to the overall performance boost. These results highlight the efficacy of GANs in data augmentation. Figure 6 presents a comparison of training and validation loss with the confusion matrix using google inceptionV3 based on CycleGAN data augmentation. 5 Conclusion The objective of our study was to identify Alzheimer's disease using MRI scans through the use of convolutional neural network models that utilized the Google Inceptionv3 CNN architecture. We conducted experiments using single-input and triple-input models and introduced generative adversarial networks (GANs) to address the challenge of limited data size in medical datasets. Our study yielded significant results that demonstrated how the incorporation of GANs can significantly improve the deep learning accuracy of Alzheimer's disease diagnosis using the ADNI1 dataset. Specifically, we utilized CycleGAN to generate images of one class from the other, thereby expanding and balancing the dataset. The experimental results showed notable improvements in categorization accuracy, where the F1 scores of the conventional model increased from 0.81 to 0.942. The implications of our study are significant since there is a shortage of large datasets in multiple medical fields. The effectiveness of using GANs for data augmentation suggests that they can significantly enhance the classification tasks across a broad range of applications. However, we identified that the current model for Alzheimer's disease diagnosis has inconsistencies due to variations in brain shape among subjects. To address this, future research could explore alternative preprocessing methods, integrate mild cognitive impairment (MCI) into the classifier, combine MRI and PET scans for more features, and include external features such as mental status tests and physical exams to improve accuracy. These improvements could lead to higher classification accuracy of Alzheimer's disease diagnoses. Declarations Author contributions: SK wrote the main manuscript text, suggested model architecture and algorithm used in the research and implemented the algorithms and provided related fgures and TA contributed to the algorithms implementation and reviewed computer aided diagnosis techniques and analysis of results. All authors reviewed the manuscript. Data availability: All the experimental data used and that support the fndings of this study are available in Alzheimer’s disease Neuroimaging Initiative (ADNI) opened database via the link https://adni.loni.usc.edu/data-samples/access-data. Competing interests : The authors declare that they have no known confict of interest associated with this publication. Funding: T here has been no signifcant fnancial support for this work that could have infuenced its outcome. References Y. Kimura et al. , “AI approach of cycle-consistent generative adversarial networks to synthesize PET images to train computer-aided diagnosis algorithm for dementia,” Ann. Nucl. Med. , vol. 34, no. 7, pp. 512–515, Jul. 2020, doi: 10.1007/S12149-020-01468-5 . J. W. Vogel et al. , “Four distinct trajectories of tau deposition identified in Alzheimer’s disease,” Nat. Med., vol. 27, no. 5, pp. 871–881, May 2021, doi: 10.1038/S41591-021-01309-6 . X. Feng, F. A. Provenzano, and S. A. Small, “A deep learning MRI approach outperforms other biomarkers of prodromal Alzheimer’s disease,” Alzheimer’s Res. Ther., vol. 14, no. 1, pp. 1–11, Dec. 2022, doi: 10.1186/S13195-022-00985-X/FIGURES/5 . T. Falk et al. , “U-Net: deep learning for cell counting, detection, and morphometry,” Nat. Methods, vol. 16, no. January, 2019, doi: 10.1038/s41592-018-0261-2 . M. S. Chong and S. Sahadevan, “Preclinical Alzheimer’s disease: diagnosis and prediction of progression,” Lancet. Neurol., vol. 4, no. 9, pp. 576–579, 2005, doi: 10.1016/S1474-4422(05)70168-X . M. Davis et al. , “Estimating Alzheimer’s Disease Progression Rates from Normal Cognition Through Mild Cognitive Impairment and Stages of Dementia,” Curr. Alzheimer Res., vol. 15, no. 8, pp. 777–788, Jan. 2018, doi: 10.2174/1567205015666180119092427 . M. Raju, M. Thirupalani, S. Vidhyabharathi, and S. Thilagavathi, “Deep Learning Based Multilevel Classification of Alzheimer’s Disease using MRI Scans,” IOP Conf. Ser. Mater. Sci. Eng., vol. 1084, no. 1, p. 012017, Mar. 2021, doi: 10.1088/1757-899X/1084/1/012017 . D. AlSaeed and S. F. Omar, “Brain MRI Analysis for Alzheimer’s Disease Diagnosis Using CNN-Based Feature Extraction and Machine Learning,” Sensors (Basel). , vol. 22, no. 8, Apr. 2022, doi: 10.3390/S22082911 . M. M. S. Fareed et al. , “ADD-Net: An Effective Deep Learning Model for Early Detection of Alzheimer Disease in MRI Scans,” IEEE Access, vol. 10, pp. 96930–96951, 2022, doi: 10.1109/ACCESS.2022.3204395 . S. Lui, X. J. Zhou, J. A. Sweeney, and Q. Gong, “Psychoradiology: The Frontier of Neuroimaging in Psychiatry,” Radiology, vol. 281, no. 2, pp. 357–372, 2016, doi: 10.1148/RADIOL.2016152149 . V. Sorin, Y. Barash, E. Konen, and E. Klang, “Creating Artificial Images for Radiology Applications Using Generative Adversarial Networks (GANs) - A Systematic Review,” Acad. Radiol. , vol. 27, no. 8, pp. 1175–1185, Aug. 2020, doi: 10.1016/J.ACRA.2019.12.024 . I. J. Goodfellow et al. , “Generative Adversarial Nets,” Adv. Neural Inf. Process. Syst. , vol. 27, 2014, Accessed: Oct. 12, 2023. [Online]. Available: http://www.github.com/goodfeli/adversarial . W. Lin et al. , “Convolutional Neural Networks-Based MRI Image Analysis for the Alzheimer’s Disease Prediction From Mild Cognitive Impairment,” Front. Neurosci. , vol. 12, no. NOV, Nov. 2018, doi: 10.3389/FNINS.2018.00777 . Y. Huang, J. Xu, Y. Zhou, T. Tong, and X. Zhuang, “Diagnosis of Alzheimer’s disease via multi-modality 3D convolutional neural network,” Front. Neurosci. , vol. 13, no. MAY, p. 448373, May 2019, doi: 10.3389/FNINS.2019.00509/BIBTEX . U. Gupta, P. K. Lam, G. Ver Steeg, and P. M. Thompson, “Improved Brain Age Estimation with Slice-based Set Networks,” Proc. - Int. Symp. Biomed. Imaging , vol. 2021-April, pp. 840–844, Feb. 2021, doi: 10.1109/ISBI48211.2021.9434081 . B. Lu et al. , “A practical Alzheimer’s disease classifier via brain imaging-based deep learning on 85,721 samples,” J. Big Data, vol. 9, no. 1, pp. 1–22, Dec. 2022, doi: 10.1186/S40537-022-00650-Y/FIGURES/5 . J. Y. Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks,” Proc. IEEE Int. Conf. Comput. Vis. , vol. 2017-October, pp. 2242–2251, Mar. 2017, doi: 10.1109/ICCV.2017.244 . “ADNI | ACCESS DATA.” https://adni.loni.usc.edu/data-samples/access-data/ (accessed Oct. 13, 2023). C. R. Jack et al. , “Prediction of AD with MRI-based hippocampal volume in mild cognitive impairment,” Neurology, vol. 52, no. 7, pp. 1397–1403, Apr. 1999, doi: 10.1212/WNL.52.7.1397 . C. Szegedy et al. , “Going Deeper with Convolutions,” Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. , vol. 07-12-June-2015, pp. 1–9, Sep. 2014, doi: 10.1109/CVPR.2015.7298594 . M. A. Morid, A. Borjali, and G. Del Fiol, “A scoping review of transfer learning research on medical image analysis using ImageNet,” Comput. Biol. Med., vol. 128, no. 408, 2021, doi: 10.1016/j.compbiomed.2020.104115 . M. Jamshidi et al. , “Artificial Intelligence and COVID-19: Deep Learning Approaches for Diagnosis and Treatment,” IEEE Access, vol. 8, pp. 109581–109595, 2020, doi: 10.1109/ACCESS.2020.3001973 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4141650","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":286973111,"identity":"8f4f6eca-18d3-451f-b1d3-38f0a612748c","order_by":0,"name":"Satish Kumar","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/klEQVRIiWNgGAWjYJCCA4wNCSCa8TGQSICIGeDTwAzXwmxMtBYGqBY2aYQWPEC+/fzBgz93pCWubT/7rLrgV12ewQHmhx8YCu7g1GJwJpnhMO+ZnMRtZ9LNbs/sO1xscIDNWILB4BluLQxALYxtFYnbDqSx3ebtOZC44QCDGVD8MG6H9T9mOPgTpOX8M7Zi3p46oBb2b3i1MNxIZjjA2wZ02I00NmaeH8xALTz4bTG48djgMG9bmvG2G8+YpWc2HC6WPMxTLJGA12GJjz/+bEuW3XY+jfFzwZ+6PL7j7Rs/fPiDx2EogLGNARxRRMQOHPwhXukoGAWjYBSMHAAAp6Jefwy9eCcAAAAASUVORK5CYII=","orcid":"","institution":"BGSB University","correspondingAuthor":true,"prefix":"","firstName":"Satish","middleName":"","lastName":"Kumar","suffix":""},{"id":286973112,"identity":"50988ce0-a3f0-4a36-a02e-f8e316fb66ef","order_by":1,"name":"Tasleem Arif","email":"","orcid":"","institution":"BGSB University","correspondingAuthor":false,"prefix":"","firstName":"Tasleem","middleName":"","lastName":"Arif","suffix":""}],"badges":[],"createdAt":"2024-03-21 07:46:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4141650/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4141650/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":55009299,"identity":"02f3aa37-ce16-40ab-9d55-510e4ee60a98","added_by":"auto","created_at":"2024-04-19 19:13:01","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":217091,"visible":true,"origin":"","legend":"\u003cp\u003eA representation of the CycleGAN architecture, which was modified for this work's research studies to diagnosis Alzheimer's disease\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/14d5d62942d833c7753c76d7.png"},{"id":55009300,"identity":"2cf87b04-fbe0-49ed-bdf7-b417518cc743","added_by":"auto","created_at":"2024-04-19 19:13:01","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":35153,"visible":true,"origin":"","legend":"\u003cp\u003eCycleGAN Loss, adapted J. Y. Zhu [17]\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/4e0f92be8b86d03bdf12c777.png"},{"id":55008400,"identity":"a17c1be1-1283-47db-aaaf-69fa0f73e0bf","added_by":"auto","created_at":"2024-04-19 19:05:01","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":114156,"visible":true,"origin":"","legend":"\u003cp\u003eThe proposed model: CycleGAN (image synthesis) + Google InceptionV3 (Diagnosis).\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/11e57ecc3e37853cd7f5b436.png"},{"id":55008404,"identity":"5cbe9449-1944-40b4-96b1-244390dfae64","added_by":"auto","created_at":"2024-04-19 19:05:01","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":40845,"visible":true,"origin":"","legend":"\u003cp\u003eCycleGAN architecture\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/c39244703ca11253884ca3a5.png"},{"id":55008401,"identity":"32dfb8fe-bc42-48f5-8e3a-d3081ba560f5","added_by":"auto","created_at":"2024-04-19 19:05:01","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":134953,"visible":true,"origin":"","legend":"\u003cp\u003eGoogle IncetionV3 architecture [21]\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/13a9b3c34f3e99643dee71ff.png"},{"id":55008403,"identity":"0d0276d9-7714-4f22-8a3b-e973e6ba06b2","added_by":"auto","created_at":"2024-04-19 19:05:01","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":103811,"visible":true,"origin":"","legend":"\u003cp\u003eResults for Google InceptionV3 with CycleGAN data augmentation\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/2c1c9e938b1ffc9fdf85f6e3.png"},{"id":63773970,"identity":"07462889-2f23-4dd3-9f90-4d9943cb4a2b","added_by":"auto","created_at":"2024-09-02 08:37:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1018685,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4141650/v1/48d38269-201c-48c5-9e70-59c070e652cf.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"CycleGAN-based Data Augmentation to Improve Generalizability Alzheimer’s Diagnosis using Deep Learning","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eAlzheimer's disease (AD) is a neurodegenerative condition primarily affecting the elderly population. According to Kimura et al. [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], it is characterized by a decline in memory, cognitive function, and behavioral function that significantly affects patients' daily activities. Studies show that the prevalence of AD is expected to triple globally by 2050, affecting over 100\u0026nbsp;million people [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. According to [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e][\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], mild cognitive impairment (MCI) is a cognitive condition that is in between the early stages of Alzheimer's disease and normal ageing. Two types of MCI are described: progressive MCI (pMCI) and stable MCI (sMCI). Patients with sMCI remain in a constant state and may even recover in some situations, while those with pMCI progress to a diagnosis of AD [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e][\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Currently, there are no clinically useful drugs for treating AD. Therefore, the major focus of treatment has shifted towards early-stage AD diagnosis[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Identifying high-risk patients and implementing tailored treatment measures to slow down the disease's progression requires differentiating between those experiencing normal cognitive decline and those with AD, sMCI, or pMCI (various stages of AD)[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e][\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Moreover, the application of image processing techniques relevant to Alzheimer's disease, such as image segmentation and denoising, can aid in the early identification of the disease.\u003c/p\u003e \u003cp\u003eIn recent years, the work of [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] serves as an example of how the field of psychoradiological research has produced useful clinical data in favour of the identification of diagnostic and therapeutic neuroimaging markers for people with psychiatric disorders. Artificial intelligence and computational psychoradiology have become more common in the diagnosis of Alzheimer's disease (AD). Still, a larger number of well-processed, high-quality images are required in order to accurately diagnose AD. Sorin et al. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], proposed the deep learning techniques used for image processing that to have some limitations. This show that the development of an efficient deep learning method that can process images is needed. A group under the direction of Goodfellow [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]unveiled an intriguing idea, the Generative Adversarial Network (GAN). In essence, GANs are deep learning models that are primarily intended to be used with images. The fundamental concept of GANs is to establish a kind of game in which the discriminator, on the one hand, tries to discern between actual and produced images, and the generator, on the other, attempts to create images. Sorin et al. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] highlighted that the interaction between the discriminator and generator is fundamental to the functioning of GANs.\u003c/p\u003e \u003cp\u003eGenerative Adversarial Networks (GANs) have the potential to significantly enhance data augmentation, especially in the medical field where there is a scarcity of available data. Unlike conventional data augmentation techniques that modify existing images, GANs can generate new, similar images from a relatively small dataset. In our research, we are exploring the possibilities of using a convolutional neural network model for classifying Alzheimer's disease. Additionally, we are assessing the feasibility of enhancing our dataset using GANs, specifically the CycleGAN design.\u003c/p\u003e"},{"header":"2 Literature Review","content":"\u003cp\u003eBrain MRI scans have been extensively utilized in previous machine learning studies to differentiate between elderly individuals suffering from Alzheimer's disease and those without cognitive problems who are of the same age. These studies often incorporate other imaging techniques or genetic data. Brain MRI scans have also been employed to distinguish between Alzheimer's patients and those suffering from other diseases or different types of dementia. Unsupervised clustering has been used to categorize patients with various forms of dementia [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This approach has been proven useful in predicting cognitive decline, such as the transition from mild cognitive impairment to Alzheimer's within a specific timeframe[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn the field of deep learning models, researchers Huang, Yechong, Xu, Jiahang, et al. have introduced a multi-task 3D CNN based on Convolutional Neural Networks (CNNs) [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. This unique CNN not only utilizes an attention mechanism to classify Alzheimer's disease but it also estimates an individual's age using an MRI scan. This task of estimating age is often referred to as the brain age problem [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Additionally, variations on this fundamental CNN design exist. For instance, hybrid Recurrent Neural Networks (RNNs) can be used to collect data from several 2D MRI slices[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. The researchers have also explored methods like pre-training and data augmentation to enhance the model's performance.\u003c/p\u003e \u003cp\u003eA recent study has utilized a state-of-the-art deep convolutional neural network (CNN) architecture, InceptionResNet-V2, to analyze a vast number of MRI scans, totaling 85,721[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. The researchers employed the pre-trained network to classify Alzheimer's disease, following a sex classification task. The results were remarkable, with accuracy rates exceeding 92% across three distinct, publicly available datasets. Furthermore, generative adversarial networks (GANs) offer significant possibilities for data augmentation and model architecture, as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Rather than merely modifying existing data in the training set, GANs can generate synthetic data by extrapolating beyond it, effectively oversampling. A novel network, CycleGAN, developed by Zhu et al. [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], can perform image-to-image translation even when the images do not directly match up. This system enables two-way translation between the source domain X and the target domain Y, with two generator networks, G1 and G2, and their corresponding partner networks, D1 and D2, participating in an adversarial training process.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWhile determining if a picture is synthetic or real, the discriminator (D) functions similarly to a decision maker. It contrasts a random authentic image from the target domain with the artificial image produced by the generator (G). It is G's responsibility to enhance the synthetic image's quality in order to trick the discriminator. After receiving an image from the source domain (x in X), the generator (G) creates a synthetic image (y\u0026thinsp;=\u0026thinsp;G(x)). After that, the discriminator (D) compares this artificial output to a randomly selected picture from the target domain (y in Y), as show in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe generator's architecture consists of three components: a decoder, a transformer (usually a residual network), and an encoder. The discriminator uses a PatchGAN model to categorise photos as artificial or real. The learning rate was initially set to 0.002 for the first 50 epochs of training, then it was progressively lowered to zero after that.\u003c/p\u003e"},{"header":"3 Methodology","content":"\u003cp\u003eThe visual representation in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e outlines the workflow of the model, which comprises of three primary phases: data collection, preprocessing utilizing CycleGAN-based augmentation, and classification via the Google InceptionV3 CNN model.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Data collection and pre-processing\u003c/h2\u003e \u003cp\u003eThe experiment is based on information provided by the Institute for Information and Communications Technology Promotion (IITP) database. The primary objective of the Alzheimer's Disease Neuroimaging Initiative (ADNI) is to determine whether mild cognitive impairment (MCI) and early Alzheimer's disease (AD) can be tracked over time through a combination of clinical and neuropsychological assessments, positron emission tomography (PET), serial magnetic resonance imaging (MRI), and other biological markers.\u003c/p\u003e \u003cp\u003eThe study employed the ADNI1 standardized MRI dataset, which was segregated into three severity-based categories: AD, MCI, and normal cognition (NC). [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The primary aim of the study was to train a neural network to differentiate between AD and NC. To achieve this, the data was converted from NIFTI files to three-dimensional NumPy arrays using nibabel. Additionally, a CSV file was extracted from the ADNI database, which included scan data, patient specifics, and the actual diagnosis labels.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe production of training samples was accomplished by slicing the three-dimensional NIFTI picture data and extracting three slices from the original image: one from the axial perspective, one from the coronal view, and one from the sagittal view. For uniformity, these slices were enlarged to 224 \u0026times; 224 from the center of each axis length. Furthermore, to separate the brain tissue from the MRI pictures and improve sample consistency, the study employed a process known as \"skull stripping,\" which involved using the deep brain library's Extractor function.\u003c/p\u003e \u003cp\u003eIn order to further refine the consistency of the data, RAS and ISO transforms and histogram normalization were applied using the TorchIO library[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. These transformations changed the orientation of the MRI picture.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 CycleGAN based data augmentation\u003c/h2\u003e \u003cp\u003eA CycleGAN model was employed in the study, comprising of two generators as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. One generator converted samples from normal cognition (NC) into Alzheimer's disease (AD) samples while the other generator did the opposite. During training, the first generator processed the real NC image to generate a fake AD image, which was then compared to a real AD image by a discriminator to calculate the GAN loss.\u003c/p\u003e \u003cp\u003eThe same training procedure was repeated with real AD images to yield another GAN loss. To build the original visuals, the fake images were passed through the second generator. A cycle consistency loss was then calculated by comparing the original NC image with its reconstruction, and repeating the process for the AD images to ensure the consistency of these reconstructions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe cycle consistency loss and GAN losses were combined to form the overall loss function, which aimed to maximize discriminators and minimize generators. The generator utilized a variety of techniques such as up-sampling, residual blocks, and down-sampling, in conjunction with the Google Inceptionv3 CNN architecture. The dataset was randomly paired after being divided into labels.\u003c/p\u003e \u003cp\u003eThree distinct CycleGAN models were trained, each targeting a different MRI slice. These models were trained for 100 epochs with a batch size of 1, using the Adam optimizer. Subsequently, the trained model generated enough samples to produce a balanced dataset. For each NC sample, an AD version was created, and vice versa. Ultimately, 480 NC and 710 AD samples of each orientation were produced, resulting in a total of 1190 pictures for each class, as shown in table 1.\u003cb\u003eTable\u0026nbsp;1.\u003c/b\u003e Dataset collection after CycleGAN\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDataset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSamples\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCycleGAN Generated\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNormal samples\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e710\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e480\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1190\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlzheimer\u0026rsquo;s Samples\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e480\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e710\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1190\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1190\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1190\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2380\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Deep CNN Classifier\u003c/h2\u003e \u003cp\u003eIn our study, we utilized the Google Inceptionv3 Convolutional Neural Network (CNN) architecture as a pre-trained model. Although the original architecture was designed for RGB images with three channels, our MRI scans are single-channel grayscale images. To adapt to the network, we replicated the same data three times along one dimension, converting them into three channels. Additionally, we modified the last layer to enable the network to function as a binary classifier.\u003c/p\u003e \u003cp\u003eWe employed a modified CNN with several inputs to effectively handle the volumetric data. The outputs from each of the three distinct Google Inceptionv3 CNN in this model are pooled and directed through fully connected layers to produce the diagnosis group at the end.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe neural network underwent training for 100 epochs with a batch size of 16 using the Adam optimizer at a learning rate of .0002. We partitioned our data into three sets: 70% of the data was utilized for training, 20% for validation, and 10% for testing purposes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Model Evaluation:\u003c/h2\u003e \u003cp\u003eIn Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e, we present a compilation of various performance measures employed in this research.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance metrics used in deep learning studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFormula\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\text{Accuracy}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{\\text{TP+TN}}{\\text{TP + TN + FP + FN }}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\text{Recall}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{\\text{TP}}{\\text{TP + FN }}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\text{Precision}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{\\text{TP}}{\\text{TP + FP }}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\text{F1-Score}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\text{2*}\\frac{\\text{Precion*Recall}}{\\text{Precion+Recall}}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4 Results and Discussion","content":"\u003cp\u003eThe data augmentation process using CycleGAN for both NC and AD samples resulted in an additional 480 and 710 samples for each orientation, respectively. These results are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e3\u003c/span\u003e, indicating that the total number of images for each class increased to 1190. We observed a significant improvement in the performance of the CNN model by employing GAN augmentation.[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The F1 score of the Google Inceptionv3 model increased by 9.4% from 0.900 to 0.942, indicating that the addition of CycleGAN greatly enhances the CNN's classification performance.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of CNN based models CycleGAN augmentation\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModels\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1 Score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInspectionV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.748\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.852\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.811\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInspectionV3\u0026thinsp;+\u0026thinsp;GAN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.918\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.880\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.921\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.900\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInspectionV3\u0026thinsp;+\u0026thinsp;CycleGAN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.922\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.934\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.942\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe synthesized images contained significant elements that are likely responsible for the beneficial effect. Moreover, the larger dataset and improved class balance in the CycleGAN-augmented data likely contributed to the overall performance boost. These results highlight the efficacy of GANs in data augmentation. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents a comparison of training and validation loss with the confusion matrix using google inceptionV3 based on CycleGAN data augmentation.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eThe objective of our study was to identify Alzheimer's disease using MRI scans through the use of convolutional neural network models that utilized the Google Inceptionv3 CNN architecture. We conducted experiments using single-input and triple-input models and introduced generative adversarial networks (GANs) to address the challenge of limited data size in medical datasets. Our study yielded significant results that demonstrated how the incorporation of GANs can significantly improve the deep learning accuracy of Alzheimer's disease diagnosis using the ADNI1 dataset. Specifically, we utilized CycleGAN to generate images of one class from the other, thereby expanding and balancing the dataset. The experimental results showed notable improvements in categorization accuracy, where the F1 scores of the conventional model increased from 0.81 to 0.942. The implications of our study are significant since there is a shortage of large datasets in multiple medical fields. The effectiveness of using GANs for data augmentation suggests that they can significantly enhance the classification tasks across a broad range of applications. However, we identified that the current model for Alzheimer's disease diagnosis has inconsistencies due to variations in brain shape among subjects. To address this, future research could explore alternative preprocessing methods, integrate mild cognitive impairment (MCI) into the classifier, combine MRI and PET scans for more features, and include external features such as mental status tests and physical exams to improve accuracy. These improvements could lead to higher classification accuracy of Alzheimer's disease diagnoses.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor contributions:\u003c/strong\u003e \u003cstrong\u003eSK\u003c/strong\u003e wrote the main manuscript text, suggested model architecture and algorithm used in the research \u0026nbsp;and implemented the algorithms and provided related fgures and \u003cstrong\u003eTA\u003c/strong\u003e contributed to the algorithms implementation and reviewed computer aided diagnosis techniques and analysis of results. All authors reviewed the manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability:\u003c/strong\u003e All the experimental data used and that support the fndings of this study are available in Alzheimer\u0026rsquo;s disease Neuroimaging Initiative (ADNI) opened database via the link https://adni.loni.usc.edu/data-samples/access-data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e: The authors declare that they have no known confict of interest associated with this publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding: T\u003c/strong\u003ehere has been no signifcant fnancial support for this work that could have infuenced its outcome.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eY. Kimura \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;AI approach of cycle-consistent generative adversarial networks to synthesize PET images to train computer-aided diagnosis algorithm for dementia,\u0026rdquo; \u003cem\u003eAnn. Nucl. Med.\u003c/em\u003e, vol. 34, no. 7, pp. 512\u0026ndash;515, Jul. 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/S12149-020-01468-5\u003c/span\u003e\u003cspan address=\"10.1007/S12149-020-01468-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ. W. Vogel \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Four distinct trajectories of tau deposition identified in Alzheimer\u0026rsquo;s disease,\u0026rdquo; Nat. Med., vol. 27, no. 5, pp. 871\u0026ndash;881, May 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/S41591-021-01309-6\u003c/span\u003e\u003cspan address=\"10.1038/S41591-021-01309-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eX. Feng, F. A. Provenzano, and S. A. Small, \u0026ldquo;A deep learning MRI approach outperforms other biomarkers of prodromal Alzheimer\u0026rsquo;s disease,\u0026rdquo; Alzheimer\u0026rsquo;s Res. Ther., vol. 14, no. 1, pp. 1\u0026ndash;11, Dec. 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/S13195-022-00985-X/FIGURES/5\u003c/span\u003e\u003cspan address=\"10.1186/S13195-022-00985-X/FIGURES/5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eT. Falk \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;U-Net: deep learning for cell counting, detection, and morphometry,\u0026rdquo; Nat. Methods, vol. 16, no. January, 2019, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41592-018-0261-2\u003c/span\u003e\u003cspan address=\"10.1038/s41592-018-0261-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. S. Chong and S. Sahadevan, \u0026ldquo;Preclinical Alzheimer\u0026rsquo;s disease: diagnosis and prediction of progression,\u0026rdquo; Lancet. Neurol., vol. 4, no. 9, pp. 576\u0026ndash;579, 2005, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S1474-4422(05)70168-X\u003c/span\u003e\u003cspan address=\"10.1016/S1474-4422(05)70168-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Davis \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Estimating Alzheimer\u0026rsquo;s Disease Progression Rates from Normal Cognition Through Mild Cognitive Impairment and Stages of Dementia,\u0026rdquo; Curr. Alzheimer Res., vol. 15, no. 8, pp. 777\u0026ndash;788, Jan. 2018, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2174/1567205015666180119092427\u003c/span\u003e\u003cspan address=\"10.2174/1567205015666180119092427\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Raju, M. Thirupalani, S. Vidhyabharathi, and S. Thilagavathi, \u0026ldquo;Deep Learning Based Multilevel Classification of Alzheimer\u0026rsquo;s Disease using MRI Scans,\u0026rdquo; IOP Conf. Ser. Mater. Sci. Eng., vol. 1084, no. 1, p. 012017, Mar. 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1088/1757-899X/1084/1/012017\u003c/span\u003e\u003cspan address=\"10.1088/1757-899X/1084/1/012017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD. AlSaeed and S. F. Omar, \u0026ldquo;Brain MRI Analysis for Alzheimer\u0026rsquo;s Disease Diagnosis Using CNN-Based Feature Extraction and Machine Learning,\u0026rdquo; \u003cem\u003eSensors (Basel).\u003c/em\u003e, vol. 22, no. 8, Apr. 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/S22082911\u003c/span\u003e\u003cspan address=\"10.3390/S22082911\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. M. S. Fareed \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;ADD-Net: An Effective Deep Learning Model for Early Detection of Alzheimer Disease in MRI Scans,\u0026rdquo; IEEE Access, vol. 10, pp. 96930\u0026ndash;96951, 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ACCESS.2022.3204395\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2022.3204395\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS. Lui, X. J. Zhou, J. A. Sweeney, and Q. Gong, \u0026ldquo;Psychoradiology: The Frontier of Neuroimaging in Psychiatry,\u0026rdquo; Radiology, vol. 281, no. 2, pp. 357\u0026ndash;372, 2016, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1148/RADIOL.2016152149\u003c/span\u003e\u003cspan address=\"10.1148/RADIOL.2016152149\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eV. Sorin, Y. Barash, E. Konen, and E. Klang, \u0026ldquo;Creating Artificial Images for Radiology Applications Using Generative Adversarial Networks (GANs) - A Systematic Review,\u0026rdquo; \u003cem\u003eAcad. Radiol.\u003c/em\u003e, vol. 27, no. 8, pp. 1175\u0026ndash;1185, Aug. 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/J.ACRA.2019.12.024\u003c/span\u003e\u003cspan address=\"10.1016/J.ACRA.2019.12.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eI. J. Goodfellow \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Generative Adversarial Nets,\u0026rdquo; \u003cem\u003eAdv. Neural Inf. Process. Syst.\u003c/em\u003e, vol. 27, 2014, Accessed: Oct. 12, 2023. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.github.com/goodfeli/adversarial\u003c/span\u003e\u003cspan address=\"http://www.github.com/goodfeli/adversarial\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eW. Lin \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Convolutional Neural Networks-Based MRI Image Analysis for the Alzheimer\u0026rsquo;s Disease Prediction From Mild Cognitive Impairment,\u0026rdquo; \u003cem\u003eFront. Neurosci.\u003c/em\u003e, vol. 12, no. NOV, Nov. 2018, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/FNINS.2018.00777\u003c/span\u003e\u003cspan address=\"10.3389/FNINS.2018.00777\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eY. Huang, J. Xu, Y. Zhou, T. Tong, and X. Zhuang, \u0026ldquo;Diagnosis of Alzheimer\u0026rsquo;s disease via multi-modality 3D convolutional neural network,\u0026rdquo; \u003cem\u003eFront. Neurosci.\u003c/em\u003e, vol. 13, no. MAY, p. 448373, May 2019, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/FNINS.2019.00509/BIBTEX\u003c/span\u003e\u003cspan address=\"10.3389/FNINS.2019.00509/BIBTEX\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eU. Gupta, P. K. Lam, G. Ver Steeg, and P. M. Thompson, \u0026ldquo;Improved Brain Age Estimation with Slice-based Set Networks,\u0026rdquo; \u003cem\u003eProc. - Int. Symp. Biomed. Imaging\u003c/em\u003e, vol. 2021-April, pp. 840\u0026ndash;844, Feb. 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ISBI48211.2021.9434081\u003c/span\u003e\u003cspan address=\"10.1109/ISBI48211.2021.9434081\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eB. Lu \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;A practical Alzheimer\u0026rsquo;s disease classifier via brain imaging-based deep learning on 85,721 samples,\u0026rdquo; J. Big Data, vol. 9, no. 1, pp. 1\u0026ndash;22, Dec. 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/S40537-022-00650-Y/FIGURES/5\u003c/span\u003e\u003cspan address=\"10.1186/S40537-022-00650-Y/FIGURES/5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ. Y. Zhu, T. Park, P. Isola, and A. A. Efros, \u0026ldquo;Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks,\u0026rdquo; \u003cem\u003eProc. IEEE Int. Conf. Comput. Vis.\u003c/em\u003e, vol. 2017-October, pp. 2242\u0026ndash;2251, Mar. 2017, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICCV.2017.244\u003c/span\u003e\u003cspan address=\"10.1109/ICCV.2017.244\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u0026ldquo;ADNI | ACCESS DATA.\u0026rdquo; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://adni.loni.usc.edu/data-samples/access-data/\u003c/span\u003e\u003cspan address=\"https://adni.loni.usc.edu/data-samples/access-data/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (accessed Oct. 13, 2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eC. R. Jack \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Prediction of AD with MRI-based hippocampal volume in mild cognitive impairment,\u0026rdquo; Neurology, vol. 52, no. 7, pp. 1397\u0026ndash;1403, Apr. 1999, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1212/WNL.52.7.1397\u003c/span\u003e\u003cspan address=\"10.1212/WNL.52.7.1397\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eC. Szegedy \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Going Deeper with Convolutions,\u0026rdquo; \u003cem\u003eProc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit.\u003c/em\u003e, vol. 07-12-June-2015, pp. 1\u0026ndash;9, Sep. 2014, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/CVPR.2015.7298594\u003c/span\u003e\u003cspan address=\"10.1109/CVPR.2015.7298594\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. A. Morid, A. Borjali, and G. Del Fiol, \u0026ldquo;A scoping review of transfer learning research on medical image analysis using ImageNet,\u0026rdquo; Comput. Biol. Med., vol. 128, no. 408, 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2020.104115\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2020.104115\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Jamshidi \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Artificial Intelligence and COVID-19: Deep Learning Approaches for Diagnosis and Treatment,\u0026rdquo; IEEE Access, vol. 8, pp. 109581\u0026ndash;109595, 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ACCESS.2020.3001973\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2020.3001973\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"CycleGAN, Data Augmentation, MRI, harmonization deep learning, Alzheimer’s disease, InceptionV3","lastPublishedDoi":"10.21203/rs.3.rs-4141650/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4141650/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAlzheimer's disease is a degenerative condition that progressively damages brain neurons, ultimately leading to dementia and death. Despite the limited number of available samples, effective diagnostic methods are crucial to diagnose Alzheimer's disease. Typically, a combination of laboratory and neuro-psychological testing is employed for diagnosis. The decrease in brain mass linked to Alzheimer's disease can be identified by MRI scans, which makes it a suitable problem for deep learning and computer vision. A precise and effective deep learning model would provide physicians with valuable support for their diagnoses. However, medical data is often challenging to obtain, and deep learning requires considerable data. To address this issue, generative adversarial networks can be useful. In this study, we proposed a CycleGAN to generate relevant synthetic images of intestinal parasites to solve the data scarcity challenge. To classify Alzheimer's disease using MRI scans, we developed convolutional neural networks based on the Google Inceptionv3 CNN architecture for this study. We attained an impressive F-1 score of 89%. Furthermore, we demonstrated the effectiveness of GANs in enhancing classification accuracy when used for data augmentation by creating samples with CycleGAN, achieving a remarkable F-1 score of 95%.\u003c/p\u003e","manuscriptTitle":"CycleGAN-based Data Augmentation to Improve Generalizability Alzheimer’s Diagnosis using Deep Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-04-19 19:04:57","doi":"10.21203/rs.3.rs-4141650/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"37ab1906-1064-4819-a981-071f6ed9331b","owner":[],"postedDate":"April 19th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-09-02T08:29:15+00:00","versionOfRecord":[],"versionCreatedAt":"2024-04-19 19:04:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4141650","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4141650","identity":"rs-4141650","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}