Unveiling New Strategies Facilitating the Implementation of Artificial Intelligence in Neuroimaging for the Early Detection of Alzheimer’s Disease

Detecting AD-related structural changes: voxel-based morphometry

Voxel-based morphometry (VBM), a structural MRI which is widely employed neuroimaging methodology enables the quantitative assessment of regional variations in brain volume and cortical thickness [23]. Studies employing VBM consistently reveal significant reductions in hippocampal volume and cortical thinning in individuals diagnosed with AD when compared to healthy matched controls. These findings correspond to the characteristic neuronal loss and atrophy observed in brain regions crucial for memory formation and retrieval [24]. The volumetric reduction of the hippocampus, serves as a prominent hallmark feature of AD pathology. Moreover, cortical thinning is evident in regions such as the entorhinal cortex, temporal lobes, and prefrontal cortex, which are critically implicated in various cognitive functions affected by AD [25]. VBM-based investigations provide valuable quantitative insights into the structural alterations associated with AD, contributing to our understanding of the disease’s neuropathology at a regional level.

In spite of the significant benefits of VBM for evaluating brain structure, this technique presents several challenges that need to be addressed. These include registration errors that can introduce inaccuracies, partial volume effects that impact voxel classification, limited spatial resolution hindering the detection of subtle changes, and the absence of temporal information for tracking disease progression. Additionally, confounding factors such as age, gender, and comorbidities can influence brain structure independently of AD, requiring careful consideration during analysis of the data obtained with VBM [26].

However, despite the challenges, VBM retains its significance as a valuable tool for investigating brain structural alterations. By addressing these limitations and synergistically combining VBM with other complementary imaging techniques and biomarkers, researchers have the potential to further advance the understanding of the underlying neurodegenerative processes in AD. This integrative approach holds promise for improving diagnostic accuracy, facilitating early detection, and enabling the development of targeted therapeutic strategies.

Deciphering microstructural changes in AD: a diffusion tensor imaging study

Diffusion tensor imaging (DTI) is a non-invasive method of neuroimaging that is typically obtained using MRI which offers valuable information about the structural health of white matter pathways within the brain. Through the analysis of both the direction and magnitude of water diffusion, DTI facilitates the generation of images and quantitative metrics that serve to characterize the organizational features of white matter fibers. Parameters such as fractional anisotropy (FA) and mean diffusivity (MD) are employed to delineate and quantify the structural characteristics of these neural pathways. This approach allows for a comprehensive assessment of white matter microarchitecture, providing valuable insights into the intricacies of neural connectivity and structural integrity in the brain.

In the context of AD, DTI has substantiated its utility as an invaluable tool for investigating white matter alterations concomitant with the pathology of the disease. This aspect is very important since AD is known to affect not only gray matter regions, but also the white matter tracts that connect them [27].

DTI studies in AD patients have consistently demonstrated specific changes in diffusion parameters. Reduced fractional anisotropy (FA) and increased mean diffusivity (MD) are commonly observed in affected individuals. FA reflects the directionality and coherence of water diffusion within white matter tracts, while MD represents the magnitude of diffusion. The reductions in FA and increases in MD observed in AD patients suggest compromised white matter integrity [28].

The observed alterations in DTI metrics are particularly prominent in key white matter tracts that connect regions affected by AD pathology. As an example, the fornix, a structure pivotal for memory function, frequently displays reduced FA and elevated MD in individuals diagnosed with AD. Similarly, the cingulum, involved in various cognitive processes, shows altered DTI metrics in AD. These changes in white matter integrity within specific tracts provide valuable insights into the structural disruption occurring in AD [29].

The DTI-derived metrics of white matter integrity have the potential to serve as biomarkers for disease progression in AD. Longitudinal studies utilizing DTI have revealed that fluctuations in FA and MD are intricately associated with the trajectory of cognitive decline. Consequently, the observed alterations in white matter integrity via DTI hold promise as a means to longitudinally monitor disease progression (like other imaging techniques) and assess the efficacy of interventions within the context of clinical trials [30]. Furthermore, DTI findings highlight the relevance of white matter alterations in AD-related cognitive impairments. The disrupted white matter integrity observed in DTI studies correlates with the severity of cognitive symptoms, including memory and executive function deficits, in AD patients [31].

Functional MRI: Investigating neural activity and connectivity in the human brain

fMRI is a robust neuroimaging modality that enables the assessment of cerebral activity and functional connectivity, providing critical insights into the functional alterations in AD [35, 36].

It allows the measurement of brain activation either at rest or in response to cognitive tasks or stimuli that engage specific brain regions and networks. It does not require the injection of contrast agents or radiation exposure, and can be repeated multiple times during longitudinal studies [18]. It provides both high temporal and spatial resolution as compared to PET; however, temporal resolution is lower than other modalities such as electroencephalography (EEG) and magnetoencephalography.

Numerous studies have investigated brain activation patterns in individuals with AD compared to those with normal cognition. These studies have consistently found decreased brain activity in parietal and hippocampal regions, which are known to be affected by the disease. Additionally, individuals with AD have shown higher activity in primary cortices, which are less affected by the disease.

Furthermore, fMRI studies have also examined the differences in activation patterns between individuals with mild cognitive impairment (MCI), a condition often considered a prodromal stages of AD and healthy controls. These studies have also identified distinct patterns of fMRI activation responses that differentiate MCI from healthy individuals [37–39].

Resting-state fMRI investigations have also played a pivotal role in unveiling disrupted patterns of functional connectivity, particularly within the default mode network (DMN), intricately involved in introspection and memory processing. Perturbations in DMN connectivity have emerged as potential biomarkers for early detection of AD, facilitating the identification of individuals at risk even prior to the manifestation of clinical symptoms [40]. The findings highlight the utility of fMRI in elucidating the neural underpinnings of AD and suggest the potential of DMN connectivity as a promising diagnostic marker for preclinical stages of the disease. The application of fMRI in AD clinical research contributes significantly to understanding of the functional alterations in the brain and may facilitate the development of early intervention strategies for the disease [41].

Concurrently, task-based fMRI paradigms have yielded significant findings by elucidating abnormal activation patterns within cognitive processes affected by AD, such as memory encoding and retrieval [42]. By employing specific cognitive tasks during fMRI acquisitions, task-based experiments have consistently revealed atypical patterns of brain activation in AD patients compared to matched healthy controls. Notably, AD patients commonly exhibit diminished activation in the hippocampus and other memory-related regions during memory encoding tasks [43]. These task-based fMRI findings provide anatomical and structural insights on the underlying functional deficits that contribute to cognitive impairments in AD.

Amyloid-β protein

The Aβ protein is recognized for its involvement in the creation of amyloid plaques observed in the brains of AD patients. Under typical physiological conditions, the generation of the Aβ protein occurs as a result of the regular processing of the amyloid-β protein precursor (AβPP) through a sequential process involving β-secretase followed by the gamma-secretase complex. The production and clearance of Aβ are tightly regulated in normal condition, ensuring a balance in its levels within the central nervous system. Aβ protein exerts influence on synaptic plasticity, neuronal development, and antioxidant activity within the brain [48, 49]. However, in AD, there is a disturbance in the equilibrium between Aβ production and clearance, leading to the accumulation of Aβ in the brain. The abnormal buildup of Aβ triggers a series of pathological processes, including inflammation and oxidative stress. As a result, the amalgamation of these pathological processes ultimately leads to neuronal dysfunction and synaptic connection loss [50].

PET imaging with radiotracers targeting Aβ, such as PiB or florbetapir, allows for the detection and quantification of Aβ deposition in the brains of individuals with AD. Increased Aβ deposition, particularly in regions like the hippocampus and cortical areas, is associated with the presence and progression of AD pathology [51]. These imaging techniques provide valuable insights into the extent of Aβ burden in the brain and assist in the diagnosis and monitoring of AD.

Tau protein

Tau is an abundant neuronal protein expressed throughout the nervous system, crucial for the promotion and stabilization of axonal microtubules. Its function is regulated by a dynamic equilibrium between a less phosphorylated state, where it associates with axonal microtubules, and a more phosphorylated state, where it becomes insoluble within the axoplasm [52]. This delicate balance is mediated by the activity of different axonal kinases and phosphatases. Perturbations in this equilibrium induce conformational changes in tau, which lead to its aggregation and altered solubility [53]. These alterations in tau structure confer resistance to cellular clearance mechanisms, including autophagy, impeding the efficient removal of tau species. In AD, abnormal accumulation of phosphorylated tau protein leads to the formation of neurofibrillary tangles. PET imaging use tau-specific radiotracers, such as flortaucipir (FTP) or AV-1451 which allows researchers to visualize and quantify the extent of tau pathology in the brain [54]. The authors discuss the development and utilization of various tau-specific PET radiotracers, such as AV-1451 (flortaucipir), THK5351, and PBB3, which selectively bind to tau deposits which enabled the visualization and measurement of tau burden, aiding in the diagnosis and monitoring of tauopathies. Tau imaging using PET was initially demonstrated with a radioactive compound called 2-(1-[6-(dimethylamino)-2-naphthyl]ethylidene)malononitrile (DDNP) labeled with fluorine-18 (¹⁸F-DDNP) [55]. This compound showed longer retention times in individuals diagnosed with AD and MCI, when compared to healthy individuals.

Another radiotracer called ¹¹C-phenyl/pyridinyl-butadienyl-benzothiazoles/benzothiazoliums (¹¹C-PBB3) also exhibited good retention in the brains of individuals with AD and other tau-related disorders thus aiding in the diagnosis and monitoring of the disease [56].

James et al., highlighted the significance of tau PET imaging as a valuable tool for non-invasive detection and quantification of tau aggregates in the living brain [57].

Another study demonstrated the potential of two tau-specific radiotracers, [¹⁸F]-THK5317 and [¹⁸F]-S16, for in vivo imaging of tau pathology in AD [58]. The study involved a small cohort of AD patients whose PET scans using both [¹⁸F]-THK5317 and [¹⁸F]-S16 radiotracers. The participants underwent clinical and cognitive assessments, including Aβ PET imaging, to confirm the diagnosis of AD and rule out other potential causes of cognitive impairment. The results demonstrated promising findings regarding the imaging capabilities of both [¹⁸F]-THK5317 and [¹⁸F]-S16. The radiotracers exhibited high binding affinity to tau aggregates, allowing for accurate visualization of tau pathology in the brain. Quantitative analysis revealed significant correlations between tracer uptake and cognitive impairment, further supporting the potential clinical utility of these imaging techniques.

Interestingly, the authors also explored the association between tau deposition and Aβ pathology, another hallmark of AD. The researchers found that tau accumulation patterns observed with [¹⁸F]-THK5317 and [¹⁸F]-S16 were consistent with the distribution of Aβ plaques, suggesting a close relationship between these two main pathological features in AD.

Glucose metabolism

Over time, researchers have extensively employed the ATN framework to classify biomarkers associated with AD, wherein amyloid (A), neurodegeneration (N), and phosphorylated tau (T), serve as its constituent elements. Well-established markers, including Aβ plaques and tau tangles, have been identified as robust indicators of AD pathology [59, 60]. However, emerging scientific investigations have shed light on the prevailing presence of impaired glucose metabolism in the brains of individuals with AD. Various comparative studies, employing age-matched cohorts of both healthy individuals and those with AD, revealed reduced glucose utilization in AD brains [61]. As a result, evaluating metabolism of glucose utilization stands as a compelling and influential imaging biomarker with significant potential for the early detection.

Under normal physiological conditions, the human brain is responsible for approximately 20% of the body’s total energy expenditure. This high metabolic demand is attributed to the intricate neural processes and cognitive functions carried out by the brain [62]. Changes in cerebral glucose metabolic rate and glucose utilization are indicative of alterations in synaptic excitability and neuronal activity. The ¹⁸FDG-PET has emerged as one of the most extensively studied PET imaging technique and has received regulatory approval as an AD biomarker [63].

A substantial number of studies demonstrate the effectiveness and utility of various imaging techniques in the delicate process of achieving an accurate diagnosis of AD or monitoring its progression. For instance, Shea et al. in their study demonstrated of the diagnostic impact of ¹⁸FDG PET and ¹¹C-PIB PET brain imaging for the diagnosis of AD and other dementias [64]. The researchers evaluated the clinical utility of these neuroimaging techniques in enhancing diagnostic accuracy and differentiating AD from other forms of dementia. A cohort of patients exhibiting cognitive impairments was included in the study. They underwent ¹⁸FDG PET and ¹¹C-PIB PET brain imaging in conjunction with standard clinical assessments. The imaging results were compared to the final clinical diagnosis to determine the impact of PET scans on diagnostic outcomes. The results of the study revealed a significant diagnostic impact of both ¹⁸FDG PET and ¹¹C-PIB PET brain imaging in the memory clinic setting for AD and other dementias. ¹⁸FDG PET provided valuable insights into brain metabolism, enabling differentiation between AD and non-AD dementias. Conversely, ¹¹C-PIB PET identified the presence of amyloid plaques, a hallmark of AD pathology, thereby contributing to the diagnosis of AD [64].

Visualizing amyloid and tau pathology: positron emission tomography imaging

Among various neuroimaging techniques, PET stands out as an important tool, harnessing the power of molecular neuroimaging to target specific molecular markers linked to the pathology of the specific neurological diseases. In the case of AD, PET imaging employs radioligands specific to Aβ and tau protein deposits, revolutionizing our understanding of disease pathogenesis [65]. Aβ PET imaging allows for the in vivo detection and quantification of amyloid plaques, a hallmark feature of AD, making it instrumental for early diagnosis and disease progression monitoring. This technique aids in distinguishing AD from other dementias and identifying individuals with preclinical AD. Tau PET imaging complements Aβ imaging by capturing the distribution of neurofibrillary tangles, the second hallmark of AD. This modality provides insights into the spatiotemporal progression of tau pathology, correlating with cognitive decline [66]. PET-based measures of Aβ and tau burden offer a comprehensive assessment of the underlying molecular processes, guiding treatment evaluation in clinical trials and facilitating the development of targeted interventions [67]. Integrating Aβ and tau PET imaging with clinical assessments enhances diagnostic accuracy, leading to personalized diagnostic approaches and therapies [68, 69].

Meanwhile, single-photon emission computed tomography (SPECT) emerges as a cornerstone in the comprehensive assessment and understanding of AD. As a non-invasive imaging technique, SPECT plays a pivotal role in evaluating cerebral blood flow and functional activity in individuals affected by AD. This capability is particularly crucial for early and accurate diagnosis, given its unique ability to offer complementary information to structural imaging methods. SPECT’s distinctive contribution lies in its power to differentiate AD from other forms of dementia, enhancing diagnostic precision and guiding clinicians in tailoring interventions to the specific needs of each patient.

Beyond its diagnostic prowess, SPECT serves as a critical asset in longitudinal studies, enabling researchers to delve into the dynamic changes in cerebral perfusion patterns over time. This longitudinal perspective not only contributes significantly to our understanding of the progressive nature of AD but also provides invaluable data for the development of therapeutic strategies and personalized patient care plans. In the evolving landscape of neuroimaging, SPECT stands as an indispensable tool, offering both accessibility and depth of insights that are vital for advancing our knowledge of AD and, ultimately, improving clinical outcomes for individuals affected by this complex neurodegenerative condition.

Neuroimaging techniques constitute indispensable tools in both clinical and research domains, affording non-invasive insights into the intricate structures and dynamic functions of the human brain. These methodologies, however, exhibit varying capabilities contingent upon their inherent spatial and temporal resolutions, magnetic field strengths, and particular structural targets [70]. In this exposition, a comparative table (Table 1) elucidate multiple neuroimaging modalities, delineating their spatial and temporal resolutions, conventional magnetic field strengths, and principal applications. A short guide which helps to choose the right brain imaging method that best suits according to the specific investigation needs [71].

Disease classification

To effectively discriminate individuals with AD or those at elevated risk of AD development, extensive research has been dedicated to classifying AD patients from individuals with MCI and healthy controls [59]. Classification endeavors in the AD domain commonly utilize diverse imaging techniques, including MRI, PET, and EEG [91]. In the past, traditional ML techniques like support vector machines, multi-layer perceptrons, and convolutional neural networks were commonly employed for AD classification. These methods typically achieved classification accuracies of approximately 90% [92]. Beyond classification accuracy, recent AI models demonstrate significant capabilities in tasks like predicting disease progression trajectories, identifying novel biomarkers for early detection, and providing personalized treatment recommendations. Highlighting the measurable performance in these additional tasks provides a more detailed perspective on the comprehensive impact of AI in the realm of AD, showcasing its potential to contribute meaningfully to various aspects of patient care and research.

In response to the significant progress in deep learning methodologies, researchers have initiated the application of these techniques to classify imaging data for the diagnosis of AD.

Progression prediction

Progression prediction refers to the ability of machine learning models to forecast how the disease will advance in individuals over time. These models utilize diverse data sources such as longitudinal neuroimaging data, genetic information, and clinical records. By analyzing this information, ML algorithms offer valuable insights into the patterns of disease progression and can identify individuals at high risk of developing AD, thereby enabling early interventions that may be more beneficial [93].

Predicting disease progression in AD encompasses two distinct tasks. The first task involves identifying individuals with MCI or normal cognition who have a higher likelihood of converting to AD [94]. For instance Rye et al., investigated the prediction of AD trajectory in individuals initially diagnosed with MCI using data from the AD Neuroimaging Initiative database [95]. The authors used a cohort study which was divided into stable MCI (n = 357) and MCI individuals progressing to AD (n = 321). Employing machine learning with cognitive function, hippocampal volume, and APOE genetic status as features, two independent classification methods consistently achieved an accuracy of around 70% across different validation procedures. Through the study it could suggested that the potential development of tools can assist clinicians in making prognostic decisions. Minhas et al. employed a decision support technique, integrating MRI-based biomarkers and neuropsychological measures, to identify those at risk of transitioning from MCI to AD. Using a novel method for disease progression, the study achieved a high classification AUC of 91.2% and 95.7% for predicting AD 6 months and 1 year ahead, respectively, with multimodal markers [96].

Another task entails predicting longitudinal AD-related scores, which is a more intricate challenge due to the availability of longitudinal measurements. Methods that involve deep learning have been employed to tackle the complexity associated with longitudinal measurements. These include conditional restricted Boltzmann machine, recurrent neural network, and ensemble models based on stacked convolutional neural network and bidirectional long short-term memory network [90].

Moreover, another study investigates the utilization of a Digital Neuro Signature (DNS) to assess longitudinal individual-level changes in AD [97]. The study focuses on the efficiency of DNS in measuring cognitive alterations over time, emphasizing its potential as a valuable tool for tracking disease progression and individual responses. The study evaluated Altoida’s DNS, a digital cognitive test, against conventional neuropsychological assessments for AD detection. DNS proved more efficient (10 min versus 45–120 min), with higher test-retest reliability and accuracy in detecting abnormal cognition. Two observational experiments with 525 participants demonstrated DNS’s consistent sensitivity in capturing individual-level cognitive changes compared to conventional assessments, especially in advanced disease stages. Dispersion differences were more pronounced with disease progression, and DNS intraindividual variability predicted conversion from MCI to AD. These findings have implications for patient monitoring, remote clinical trials, and timely interventions in AD management [97].

Image segmentation

Image segmentation using ML algorithms plays a crucial role in accurately delineating brain regions affected by AD pathology. By precisely segmenting brain images, ML algorithms enable quantitative assessment and monitoring of disease-related changes in neuroimaging data [98].

The process of image segmentation involves dividing an image into distinct regions or objects. In the context of AD, the aim is to identify and delineate specific brain regions that are indicative of AD pathology, such as areas of atrophy, the presence of amyloid plaques, or alterations in connectivity patterns [99]. ML algorithms are trained on labeled datasets, where experts have manually annotated the regions of interest.

Once trained, ML algorithms can automatically segment brain images with a high degree of accuracy and consistency, eliminating the subjectivity and potential inter-observer variability associated with manual segmentation. This precise image segmentation allows for quantitative analysis of disease-related changes over time and provides valuable insights into the progression and severity of AD [100].

Accurate segmentation of brain regions affected by AD pathology facilitates various applications in AD research and clinical practice. For instance, it enables the measurement of specific biomarkers associated with disease progression, such as hippocampal volume or amyloid plaque burden. This quantitative information can aid in the early detection of AD, tracking disease progression, and evaluating treatment efficacy [101].

Furthermore, image segmentation assists in characterizing disease subtypes by identifying specific patterns of neurodegeneration or identifying regions of interest that are predictive of disease conversion from MCI to AD. It also facilitates the assessment of treatment response by quantifying changes in segmented regions following therapeutic interventions.

Thus, AI methods present an opportunity to uncover novel insights and biomarkers in neuroimaging. Through harnessing the computational power of deep learning, researchers are being able to identify previously unidentified imaging markers and unveil new imaging signatures that are associated with disease progression, subtypes, or treatment response.

Despite the significant advancements in the integration of AI and deep learning for neuroimaging, several challenges persist. The availability of large and diverse datasets, encompassing comprehensive neuroimaging features and associated clinical information, remains a critical obstacle due to time, cost, and ethical considerations related to patient privacy. Another significant challenge lies in the interpretability and transparency of AI models, as deep learning algorithms often operate as opaque black boxes, necessitating the development of interpretable models for establishing trust and acceptance. Standardization and harmonization of imaging protocols and analysis methods across different sites and scanners present difficulties, as variations can significantly impact the performance and generalizability of AI models. Furthermore, extensive validation and rigorous evaluation through robust clinical studies are imperative to assess the accuracy and clinical utility of AI-driven neuroimaging techniques, while simultaneously addressing regulatory and ethical concerns surrounding model interpretability, patient consent, and potential biases.