{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        print(os.path.join(dirname, filename))\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Alzheimer's disease and multimodal AI models\n1.\tWhat is Alzheimer's disease?\nAlzheimer's disease (AD) is an irreversible, chronic, progressive, and complex disease driven mainly cause by neural loss due to the accumulation of intracellular neurofibrillary tangles and amyloid β (Aβ) aggregates in the brain [1][2]. It typically manifests after the age of 60 and is accompanied by a decline in doing personal daily activities, personality alterations, and impairment of cognitive function [3][4]. As the population ages, the prevalence of dementia is expected to triple globally by 2050 [5]. If diagnosing AD during the pre-symptomatic stages becomes more reliable, quicker, more successful therapeutic and disease follow-up methods can be achieved [6].\nMagnetic resonance imaging (MRI) is critical in diagnosing AD through clinical reasoning [7]. Furthermore, identifying biomarkers for early AD detection is essential; however, measuring these biomarkers in the blood is challenging due to their low concentrations. Hence, developing a technical modality like immunomagnetic reduction is vital for detecting Aβ1–42 and tau proteins [9]. Among these methodologies, Arterial Spin Labeling (ASL) has emerged as a salient sub-method of MRI with remarkable potential in elucidating the underlying mechanisms of AD [7][8]. ASL-MRI, a non-invasive neuroimaging modality, offers unique advantages in assessing cerebral blood flow (CBF), providing new insights into the early stages of AD pathology without the need for exposure to radioactivity or contrast agents [9]. By using the inherent capabilities of ASL-MRI, researchers, and clinicians gain unprecedented insights into the CBF abnormalities, often preceding the manifestation of overt cognitive impairments and structural brain atrophy commonly associated with AD [7]. The observed reduction in CBF, detected through the ASL-MRI, is an early indicator of the accumulation of pathogenic proteinaceous toxins, such as Aβ and phosphorylated tau (p-tau), recognized as crucial contributors to AD etiology. By early identification of CBF patterns, ASL-MRI facilitates prompt intervention and treatment, potentially mitigating the disease's devastating impact [7]–[9].\nTraditional methods of diagnosing Alzheimer's disease involve brain-imaging tests and the examination of molecular biomarkers, such as Aβ42, p-tau, and total tau (t-tau) levels in cerebrospinal fluid (CSF) [10]. However, these methods are often invasive or expensive. Consequently, using less invasive and more accessible biomarkers becomes vital for effective screening and early diagnosis of Alzheimer's disease [10][11]. In this context, blood-based biomarkers offer promising prospects for addressing this challenge [11]. AlzoSure® Predict is a noteworthy blood-based biomarker, offering a rapid and minimally invasive blood test capable of predicting AD at least six years before the manifestation of clinical symptoms [12]. The AlzoSure® Predict method is utilized for measuring U-p53AZ levels in plasma samples of individuals at different stages of AD, even the preclinical stage. By employing the AlzoSure® Predict method, it is possible to obtain reliable and accurate measurements of U-p53AZ, which could have diagnostic and monitoring implications for tracking the progression of AD [13].\nArtificial intelligence models, specifically image and text processing models such as Vision-language pre-training (VLP) models, have been demonstrated to be effective in many applications and could aid and help with early detection, disease Progression Tracking, differential diagnosis, and predictive modeling of Alzheimer's disease.[14]\nIn conclusion, early detection of Alzheimer's disease is essential for implementing timely interventions and developing effective treatment strategies to mitigate cognitive decline and enhance individuals' overall well-being. These techniques can predict the onset of Alzheimer's disease at an early stage, even before the manifestation of clinical symptoms, facilitating early interventions and potentially improving prognostic detection, prompt action, and patients' overall quality of life.\n\n\n2.\tWhat is ASL?\nMagnetic resonance imaging (MRI) is a medical imaging technique that uses a magnetic field and computer-generated radio waves to create detailed images of the organs and tissues in the human body.\nArterial spin labeling(ASL) is one of the MIR sub-method. Using a freely diffusible intrinsic tracer, we use this method to measure tissue perfusion and Cerebral blood flow (CBF) in the brain.to get the magnetically labeled arterial blood below the imaging slab in the ALS image, the MRI technologists or radiology technologists utilize the water molecules circulating in the blood vessels of the neck, which then goes to the brain and, using a radiofrequency pulse, track the blood water as it circulates throughout the brain. A ' label ' image is captured after a specific time in microseconds (enough time to allow the blood to circulate through the brain vessels).\nOnce the images are taken, the medical radiologists will use them to diagnose diseases and conditions using medical imaging techniques. The analysis and the report of the radiologists finding are usually referred to the patient's designated physician, who then uses these pieces of information to make a diagnosis or determine further treatment plans.[15]\n\n3.\tWhat is AlzoSure®?\nAlzoSure® is a simple, non-invasive blood test that could predict Alzheimer's disease up to 6 years before mild cognitive impairment (MCI) happens in patients and before the clinical symptoms start to be identifiable.[16][17]\n\n4.\tAlzheimer's Disease Neuroimaging Initiative (ADNI)\n\nUsing ASL image and AlzoSure®, we could start training the artificial intelligence models that help the medical staff team identify Alzheimer's disease timely and have a proper responding treatment strategy to improve patient quality of life. The Data needed and used in the further part of this paper is from The Alzheimer's Disease Neuroimaging Initiative (ADNI).ADNI provides researchers with various reliable data that can be used as free access data sources. I mostly used ASL images and AlzoSure® blood test results, but ADNI provided many other data types that could be used as main or supplementary data.[18]\n\n5.\tmultimodal models\nmultimodal models are generally called artificial intelligence models that can process text and images for prediction tasks. One notable and exemplary one is vision and language models (VLMs). VLMs try to understand and generate meaningful connections between visual and textual information provided to them. These models usually design with deep neural network architectures. For image processing tasks, you could often use convolutional neural networks(CNN); for natural language processing, one can use the recurrent neural network (RRN) or transformer models. a VLM model typically integrates and processes image and text data to extract information and perform prediction.\nVLMs require large datasets, and the training process involves image and text representation so the model can realize semantic relationships between two modalities.\nSome examples of VLMs are ViT (Vision Transformer), LXMERT (Cross-Modal Transformer), and UNITER (Universal Image-Text Representation Learning). These models have demonstrated impressive performance on various multimodal prediction tasks, driven by advancements in deep learning and large-scale datasets.\nRemember that developing and training these models need substantial computational resources and labeled multimodal datasets. ADNI provides you with data; you could always use colab and Kaggle to test drive the models.[14]–[16], [19]\n\n6.\tadvantages and this advantage of different multimodal AI models\nDifferent multimodal AL models have advantages and disadvantages based on the architecture, training method, and applications, but I tried to summarize them.[20]\n\nAdvantages:\nEnhance performance: as you provide the models with text and image data, models usually excel unimodal by far and can extract richer representations and improve overall performance. The fusion of information also can lead to improved accuracy, robustness, and generalization across tasks.\nComplementary Information and Contextual Understanding: multimodal's can leverage text and image data and have a better and more comprehensive understanding of underlying patterns or phenomena. They are also better at comprehending the context and semantic relationship between text and images by jointly processing both modalities.\nThese models can infer connections that might not be apparent when considering each modality independently.\nImprove interpretability: in some cases. In some instances, multimodal models can enhance interpretability by explicitly relating visual and textual elements.\nUsing a multimodal model instead of a unimodal model can aid in understanding the reasoning or decision-making process of the model; this is particularly important in our task in hand, which is the very sensitive domain of healthcare. [18], [21]\n\nDisadvantages:\nData collection and labeling: building multimodal datasets is time-consuming and challenging. In our case, you need to acquire and annotate a set of datasets, and different medical experts should do the labeling task; hence the available data and data sources are quite limited, and limited data may cause your model to the lake many variants and rare condition data which will impact model accuracy directly.\nModel complexity: multimodal models are more complex than unimodal models since multimodal models need to process and integrate multiple modalities. These complexities require more computational and training time and extended resources. Training processes also need careful design choices and techniques so the model can effectively merge text and images. Finding a proper loss function and learning strategies should be considered before the \nInteractions between different modalities could happen.\nDomain-specific Adaptation: Multimodal models trained on generic datasets may not perform optimally in specific domains or applications. Adapting these models to particular tasks or domains often requires additional fine-tuning or transfer learning approaches or retraining and reruns.[18], [21]\n\n\n7.\tDifferent VMLs applications\n\nImage captioning: in our in-hand task, we can use the model to caption our ASL image to identify Alzheimer's disease. It could also provide with regain of the brain started to degenerate.\n\nVisual Question Answering (VQA) and Visual Dialogue: the model can be set up as a web application by taking the necessary data (image and blood sample reports) and answering questions, or you can add a Visual Dialogue to the architecture, and it could be enabled to dialogue systems that comprehend and respond to multimodal inputs.\n\nCross-Modal Retrieval: model can also find relevant images to textual information given an input of text or image. For example, In our case, this model can be used to find similar patients who have missing data and could be helpful to medical staff to have better references. [22], [23]\n\n\n8.\tMultimodal AI models examples associated with Alzheimer's Disease \n\nSeveral multimodal AI models have been developed for Alzheimer's disease (AD) research and detection. Here I mention some of them.\n\nADNet: ADNet is a multimodal model that integrates MIR data and genetic information. It uses deep learning techniques with a graph convolutional neural network to predict AD progress, and it could also identify biomarkers. You could use AD with ASL and AlzoSure® data.[24]\n\nMIRIAD: MIRIAD is fabricated of multimodal imaging in Alzheimer's disease. MIRIAD is a multimodal dataset that can combine MRI scans and cerebrospinal fluid biomarker (CBF) measurements with clinical assessments of AD patients. Many multimodal models utilize MIRIAD advanced dataset to create other models which could be used in studying AD progression, classifying AD subtypes, and predicting patient neurodegenerative disease advancement from mild cognitive impairment (MCI) to AD.[25]\n\nMRI_Text Integeration: some studies integrated MRI scans with textual clinical notes or neuropsychological test results. Multimodal models that combine features from both modalities can use to detect AD, predict disease progression, and identify relevant imaging biomarkers with data gathered from integrated MRI scans.[26]\n\nFusion of imaging and genetic data: Multimodal models that fuse genetic data (gene expression profiles) with neuroimaging data have been employed to identify genetic risk factors for AD and study the relationships between genetic variants and imaging features associated with the disease.[27]\n\nThe last category is not text and image multimodal data; it uses PET and MRI images as input.\n\nMultimodal Deep Learning with PET and MRI: In AD research, Positron emission tomography (PET) and MRI are commonly used imaging modalities. Multimodal deep learning models that integrate features from both PET and MRI scans have been developed to improve AD diagnosis, predict disease progression, and identify imaging biomarkers that correlate with clinical outcomes.[28], [29]\n\nCollaborations between researchers, clinicians, and data scientists are crucial to collect, annotating, and analyzing such datasets and pre-train models, enabling the development and Validation of robust multimodal models for AD detection, prediction, and personalized care.\n \n\n9.\tpre-developed models and approaches used for each application in the context of Alzheimer's disease (AD)\n\nSome specific pre-train artificial intelligence models can be used regarding Alzheimer's disease. I made a list of six different task categories that these pre-train multimodal models could be used.\n\nMultimodal data fusion: there are two pre-train models in this area. First, the FusionNet and second other Multi Deep Learning Approaches. FusuionNet is a deep learning model that uses fusion and combined multimodal neuroimaging data, such as MRI and PET scans, with clinical and genetic data. The second pre-train models in this area are custom-built multimodal architectures that combine convolutional neural networks (CNNs) for imaging data and recurrent neural networks (RNNs) or transformers for textual or clinical data fusion. You can find these customer designs in paper or Git Hub and use them in your models.[30]\n\nDisease Diagnosis and Classification: same as the first area, there are also two pre-train models.\nADNI-MLP, which is Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and has been used with a multi-layer perceptron (MLP) model to classify AD patients and healthy controls based on multimodal features in many studies. Also, there are Multimodal Deep Neural Network solutions. Deep learning architectures like convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are commonly used for disease classification tasks by integrating information from different modalities and making reusable custom architectures.[31]\n\nDisease Progression and Prediction:\nLongitudinal Multimodal Deep Learning and Event-Based Models are two models that can be used in these areas. Longitudinal multimodal models leverage recurrent neural networks (RNNs) or transformers to analyze sequential imaging and clinical data, enabling the prediction of AD progression and conversion from MCI to AD, and Event-based models utilize event-based learning algorithms to track changes and detect transitions in disease progression, considering multimodal inputs and could be used for personalized treatment strategies.[32]\n\nImaging Biomarker Identification: this one can use Voxel-Based Analysis or Deep Learning-based Feature Extraction methods. Voxel-based analysis methods, such as voxel-based morphometry (VBM), are often employed to identify imaging biomarkers by analyzing structural or functional changes across multimodal neuroimaging data and Deep learning models, including CNNs or autoencoders, are used to extract discriminative features from multimodal imaging data for biomarker identification.[32]\n\nGenetic and Imaging Associations: we have Polygenic Risk Scores (PRS) models, and Imaging Genomics approaches in this area. PRS models combine genetic information, such as SNPs, with multimodal neuroimaging data to identify associations between genetic risk factors and imaging features in AD, and Imaging genomics approaches integrate genetic variants and neuroimaging data to identify genetic markers associated with AD, often utilizing machine learning algorithms like linear regression or random forests.[33]\n\nTreatment Response and Outcome Prediction: Longitudinal Multimodal and Reinforcement Learning-based Models are usually used for these tasks. Longitudinal multimodal models trained on imaging, clinical, and genetic data can predict treatment response and prognosticate disease outcomes by considering changes over time. Reinforcement learning approaches can be employed to optimize treatment strategies based on multimodal patient data, aiming to predict the efficacy of different interventions.[34]\nThe mentioned area are some of the available options, and the field of multimodal AI models in the AD domain is fast growing.in this field, collaboration between researchers, clinicians, and data scientists is crucial to achieving a high standard.\n\n10.\tBenefits and impact\n\nMultimodal AI models applied in the context of Alzheimer's disease (AD) offer several benefits and significantly impact various aspects of AD research, diagnosis, and patient care. Here are some key benefits and impacts of using multimodal AI models in AD are:\nImproved accuracy and early detection, Enhanced understanding of disease mechanisms, Personalized treatment, care, and medicine, Biomarker discovery and development, Prognostication and disease progression tracking, Research efficiency and cost-effectiveness, and Clinical decision support.\nUsing multimodal AI models in AD research and clinical practice can transform our understanding of the disease, improve diagnostic accuracy, enable personalized care, and facilitate the development of targeted interventions. However, it is important to validate and interpret the results of these models in conjunction with clinical expertise to ensure their safe and effective deployment in real-world settings.\n\n11.\tConcerns and Limitations \n\nWhile multimodal AI models applied in the context of Alzheimer's disease (AD) offer numerous benefits, some important concerns and limitations should be considered. Here are some key considerations and constraints associated with the use of multimodal AI models in AD:\nData availability and quality, Interpretability and explainability, Generalization, and external Validation, Ethical and privacy concerns, Bias and disparities and morals, Clinical relevance and utility, Validation and regulatory considerations, and Human-AI collaboration. Addressing these concerns and limitations requires close cooperation between researchers, clinicians, policymakers, and regulatory bodies. Striving for transparency, data sharing, standardized protocols, interpretability, and ethical considerations can help mitigate these challenges and ensure the responsible and beneficial application of multimodal AI models in AD research and clinical practice.\n\n12.\tThe future of multimodal in AD\n\nThe future of multimodal AI in Alzheimer's disease (AD) holds great promise for advancing our understanding of the disease, improving diagnosis and prediction, and enabling personalized interventions. Here are some potential developments and directions for the future of multimodal AI in AD: Enhanced Integration of data modalities, Longitudinal monitoring and disease trajectories, Early detection and preclinical stages, improved accuracy and generalizability, Explainable and interpretable models, Personalized treatment and precision medicine, Integration into clinical workflows, Drug development, and clinical trials, Public health and population studies and Collaborative research and data sharing.\nThe future of multimodal AI in AD holds tremendous potential to transform our approach to diagnosis, treatment, and prevention. Continued technological advancements, data availability, and interdisciplinary collaborations will drive these developments, ultimately leading to improved patient outcomes and a deeper understanding of AD.\nAs multimodal AI continues to advance and become more integrated into healthcare systems, it has the potential to revolutionize AD care, improve outcomes, and guide decision-making processes. However, careful implementation, Validation, and close collaboration between AI developers, healthcare professionals, and regulatory bodies are crucial to ensure the responsible and ethical deployment of multimodal AI in AD healthcare settings.\n\n\n13.\tConclusion\n\nIn conclusion, multimodal AI models applied in the context of Alzheimer's disease (AD) hold significant promise for advancing our understanding, diagnosis, and management of the disease. These models provide a comprehensive and holistic view of AD pathology, progression, and treatment response by integrating diverse data modalities such as imaging, genetics, biomarkers, and clinical assessments. As a fast-growing field, some concerns and limitations need to be addressed, but despite these challenges, the future of multimodal AI in AD within the healthcare industry is promising. Advancements in technology, data sharing, and interdisciplinary collaborations will drive the development and adoption of these models, ultimately leading to improved patient outcomes, personalized care, and a deeper understanding of AD. Continued research, Validation, and regulatory efforts will be key in realizing the full potential of multimodal AI in tackling the complex challenges posed by Alzheimer's disease.\n\n14. references\n\n[1]\tM. O. Hurtado, I. Kohler, and E. C. M. De Lange, \"Next-generation biomarker discovery in Alzheimer's disease using metabolomics–from animal to human studies,\" Bioanalysis, vol. 10, no. 18, pp. 1525–1546, 2018.\n[2]\tM. Zvěřová, \"Clinical aspects of Alzheimer's disease,\" Clin. Biochem., vol. 72, pp. 3–6, 2019.\n[3]\tM. Beriashvili, M. Gigiadze, and G. Gorgadze, \"THE ROLE OF ENVIRONMENTAL AIR POLLUTION IN THE DEVELOPMENT OF ALZHEIMER'S DISEASE,\" Jr. Res., vol. 1, no. 1, pp. 106–114, 2023.\n[4]\tM. Hassan et al., \"Computational modeling and biomarker studies of pharmacological treatment of Alzheimer's disease,\" Mol. Med. Rep., vol. 18, no. 1, pp. 639–655, 2018.\n[5]\tP. Scheltens et al., \"Alzheimer's disease,\" Lancet, vol. 388, no. 10043, pp. 505–517, 2016.\n[6]\tA. M. Kotecha, A. D. C. Corrêa, K. M. Fisher, and J. V Rushworth, \"Olfactory dysfunction as a global biomarker for sniffing out Alzheimer's disease: a meta-analysis,\" Biosensors, vol. 8, no. 2, p. 41, 2018.\n[7]\tD. A. Wolk and J. A. Detre, \"Arterial spin labeling MRI: an emerging biomarker for Alzheimer's disease and other neurodegenerative conditions,\" Curr. Opin. Neurol., vol. 25, no. 4, p. 421, 2012.\n[8]\tT. Mutoh, K. Eguchi, S. Yamamoto, N. Yasui, and Y. Taki, \"Arterial Spin Labeling Magnetic Resonance Imaging in the Assessment of  Non-Convulsive Status Epilepticus in Alzheimer's Disease: A Report of Two Cases.,\" Am. J. Case Rep., vol. 20, pp. 1883–1887, Dec. 2019, doi: 10.12659/AJCR.919938.\n[9]\tD. C. Alsop et al., \"Recommended implementation of arterial spin-labeled perfusion MRI for clinical  applications: A consensus of the ISMRM perfusion study group and the European consortium for ASL in dementia.,\" Magn. Reson. Med., vol. 73, no. 1, pp. 102–116, Jan. 2015, doi: 10.1002/mrm.25197.\n[10]\tS. Gunes, Y. Aizawa, T. Sugashi, M. Sugimoto, and P. P. Rodrigues, \"Biomarkers for Alzheimer’s Disease in the Current State: A Narrative Review,\" International Journal of Molecular Sciences, vol. 23, no. 9. 2022, doi: 10.3390/ijms23094962.\n[11]\tC. E. Teunissen et al., \"Blood-based biomarkers for Alzheimer's disease: towards clinical implementation.,\" Lancet. Neurol., vol. 21, no. 1, pp. 66–77, Jan. 2022, doi: 10.1016/S1474-4422(21)00361-6.\n[12]\tS. Piccirella et al., \"AlzoSure® Predict, a simple minimally-invasive blood test to predict the early onset of Alzheimer's disease before the manifestation of clinical symptoms,\" Alzheimer's Dement., vol. 18, no. S6, p. e069262, Dec. 2022, doi: https://doi.org/10.1002/alz.069262.\n[13]\tA. S. Piccirella, D. srl, and I. Isb, \"Title Discovery of simple blood-based test to predict early onset of Alzheimer's using standard Clinical Mass Spectrometry platforms,\" no. January, 2021, doi: 10.20944/preprints202101.0530.v1.\n[14]\tQ. Chen, X. Hu, Z. Wang, and Y. Hong, \"MedBLIP: Bootstrapping Language-Image Pre-training from 3D Medical Images and Texts,\" 2023, [Online]. Available: http://arxiv.org/abs/2305.10799.\n[15]\tW. Zheng, B. Cui, Y. Han, H. Song, K. Li, and Y. He, \"Disrupted Regional Cerebral Blood Flow , Functional Activity and Connectivity in Alzheimer' s Disease : A Combined ASL Perfusion and Resting State fMRI Study,\" vol. 13, no. July, pp. 1–12, 2019, doi: 10.3389/fnins.2019.00738.\n[16]\tProfessor John Gallacher, \"Assessing the prospective utility of AlzoSure® Predict in a UK clinical research cohorts to assess risk of Alzheimer's DiseaseNo Title,\" university of oxford, 2023. https://www.medsci.ox.ac.uk/study/graduateschool/mrcdtp/icase-2023/assessing-the-prospective-utility-of-alzosure-r-predict-in-a-uk-clinical-research-cohorts-to-assess-risk-of-alzheimer2019s-disease.\n[17]\t\"A Conformational Variant of p53 (U-p53,\" vol. 3, no. 9, pp. 469–479, 2022.\n[18]\tA. Vasanthakumar et al., \"Harnessing peripheral DNA methylation differences in the Alzheimer' s Disease Neuroimaging Initiative ( ADNI ) to reveal novel biomarkers of disease,\" pp. 1–11, 2020.\n[19]\tC. E. Wisely et al., \"Convolutional neural network to identify symptomatic Alzheimer' s disease using multimodal retinal imaging,\" pp. 388–395, 2022, doi: 10.1136/bjophthalmol-2020-317659.\n[20]\tM. Velazquez and Y. Lee, \"Multimodal ensemble model for Alzheimer's disease conversion prediction from Early Mild Cognitive Impairment subjects,\" Comput. Biol. Med., vol. 151, p. 106201, 2022, doi: https://doi.org/10.1016/j.compbiomed.2022.106201.\n[21]\tJ. Venugopalan, \"Multimodal deep learning models for early detection of Alzheimer' s disease stage,\" Sci. Rep., pp. 1–13, 2021, doi: 10.1038/s41598-020-74399-w.\n[22]\tJ. An, \"Using CatBoost and Other Supervised Machine Learning Algorithms to Predict Alzheimer's Disease,\" in 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA), 2022, pp. 1732–1739, doi: 10.1109/ICMLA55696.2022.00265.\n[23]\tC. V. M. L. Jie, V. Treyer, and R. Schibli, \"Tauvid TM : The First FDA-Approved PET Tracer for Imaging Tau Pathology in Alzheimer' s Disease,\" 2021.\n[24]\tC. Fabrizio, A. Termine, C. Caltagirone, and G. Sancesario, “Artificial Intelligence for Alzheimer ’ s Disease : Promise or Challenge ?,” pp. 1–19, 2021.\n[25]\tA. G. Sánchez-reyna et al., \"Multimodal Early Alzheimer' s Detection , a Genetic Algorithm Approach with Support Vector Machines,\" pp. 1–19, 2021.\n[26]\tS. Jahan et al., \"Explainable AI-based Alzheimer' s Prediction and Management Using Multimodal Data,\" no. March, pp. 1–16, 2022, doi: 10.20944/preprints202203.0214.v1.\n[27]\tD. Chen, L. Zhang, and C. Ma, \"A Multimodal Diagnosis Predictive Model of Alzheimer's Disease with Few-shot Learning,\" in 2020 International Conference on Public Health and Data Science (ICPHDS), 2020, pp. 273–277, doi: 10.1109/ICPHDS51617.2020.00060.\n[28]\tQ. Zhang, Q. Du, and G. Liu, \"A whole-process interpretable and multimodal deep reinforcement learning for diagnosis and analysis of Alzheimer's disease∗,\" J. Neural Eng., vol. 18, no. 6, p. 66032, Dec. 2021, doi: 10.1088/1741-2552/ac37cc.\n[29]\tS. Qiu et al., \"dementia assessment,\" no. 2022, pp. 1–17, doi: 10.1038/s41467-022-31037-5.\n[30]\tC. Fuentes, M. Porcheron, and J. E. Fischer, RoboClean : Contextual Language Grounding for Human-Robot Interactions in Specialised Low-Resource Environments, vol. 1, no. 1. Association for Computing Machinery.\n[31]\tM. Khalil, \"Brain Predictors of Negative Symptoms and Functional Outcomes in First-Episode Psychosis: A Multimodal Connectivity Approach,\" no. April, pp. 1–19, 2022.\n[32]\tA. E. Id, Ø. Mette, T. Id, O. Emil, and A. Id, \"PLOS DIGITAL HEALTH Transfer learning for non-image data in clinical research : A scoping review,\" pp. 1–22, 2022, doi: 10.1371/journal.pdig.0000014.\n[33]\tE. Boumpa, A. Gkogkidis, I. Charalampou, and A. Ntaliani, \"An Acoustic-Based Smart Home System for People Suffering from Dementia,\" pp. 1–20, 2019, doi: 10.3390/technologies7010029.\n[34]\tS. Iddi, D. Li, P. S. Aisen, M. S. Rafii, W. K. Thompson, and M. C. Donohue, \"Predicting the course of Alzheimer' s progression,\" Brain Informatics, 2019, doi: 10.1186/s40708-019-0099-0.\n\n","metadata":{}}]}