The new era of artificial intelligence in neuroradiology: current research and promising tools

A nova era da inteligência artificial em neurorradiologia: pesquisa atual e ferramentas promissoras

Abstract

Radiology has a number of characteristics that make it an especially suitable medical discipline for early artificial intelligence (AI) adoption. These include having a well-established digital workflow, standardized protocols for image storage, and numerous well-defined interpretive activities. The more than 200 commercial radiologic AI-based products recently approved by the Food and Drug Administration (FDA) to assist radiologists in a number of narrow image-analysis tasks such as image enhancement, workflow triage, and quantification, corroborate this observation. However, in order to leverage AI to boost efficacy and efficiency, and to overcome substantial obstacles to widespread successful clinical use of these products, radiologists should become familiarized with the emerging applications in their particular areas of expertise. In light of this, in this article we survey the existing literature on the application of AI-based techniques in neuroradiology, focusing on conditions such as vascular diseases, epilepsy, and demyelinating and neurodegenerative conditions. We also introduce some of the algorithms behind the applications, briefly discuss a few of the challenges of generalization in the use of AI models in neuroradiology, and skate over the most relevant commercially available solutions adopted in clinical practice. If well designed, AI algorithms have the potential to radically improve radiology, strengthening image analysis, enhancing the value of quantitative imaging techniques, and mitigating diagnostic errors.

Keywords
Artificial Intelligence; Deep Learning; Machine Learning; Neuroradiology

Resumo

A radiologia tem uma série de características que a torna uma disciplina médica especialmente adequada à adoção precoce da inteligência artificial (IA), incluindo um fluxo de trabalho digital bem estabelecido, protocolos padronizados para armazenamento de imagens e inúmeras atividades interpretativas bem definidas. Tal adequação é corroborada pelos mais de 200 produtos radiológicos comerciais baseados em IA recentemente aprovados pelo Food and Drug Administration (FDA) para auxiliar os radiologistas em uma série de tarefas restritas de análise de imagens, como quantificação, triagem de fluxo de trabalho e aprimoramento da qualidade das imagens. Entretanto, para o aumento da eficácia e eficiência da IA, além de uma utilização clínica bem-sucedida dos produtos que utilizam essa tecnologia, os radiologistas devem estar atualizados com as aplicações em suas áreas específicas de atuação. Assim, neste artigo, pesquisamos na literatura existente aplicações baseadas em IA em neurorradiologia, mais especificamente em condições como doenças vasculares, epilepsia, condições desmielinizantes e neurodegenerativas. Também abordamos os principais algoritmos por trás de tais aplicações, discutimos alguns dos desafios na generalização no uso desses modelos e introduzimos as soluções comercialmente disponíveis mais relevantes adotadas na prática clínica. Se cautelosamente desenvolvidos, os algoritmos de IA têm o potencial de melhorar radicalmente a radiologia, aperfeiçoando a análise de imagens, aumentando o valor das técnicas de imagem quantitativas e mitigando erros de diagnóstico.

Palavras-chave
Inteligência Artificial; Aprendizado Profundo; Aprendizado de Máquina; Neurorradiologia

INTRODUCTION

Artificial intelligence (AI) has emerged as a promising tool in scientific research and medical healthcare. Machine learning (ML) and deep learning (DL) have offered means of evaluation of large datasets with intense computational developments that may provide considerate insights and new achievements in the diagnosis, prognosis, clinical, and surgical management of several neurological diseases.¹-³ Technographical reviews have demonstrated that the usage of technology may support, extend, and, to a lesser extent, replace some specific tasks performed by neuroradiologists.⁴ It is of paramount importance to understand the potential contributions and limitations of AI in neuroradiology to embrace the available techniques and foster the development of useful tools that may ultimately impact clinical practice and patient management. Herein, our focus is to review the state-of-the-art AI translational research in the field of cerebrovascular diseases, demyelinating disorders, epilepsy, and neurodegenerative conditions.

AI IN CEREBROVASCULAR DISEASES

Cerebrovascular diseases (CVDs) are a major cause of death and disability globally, frequently having devastating effects on patients and their families and a considerable negative impact on the healthcare system and the economy.⁵ For the purpose of simplicity, we will adopt the clinical outcome-based classification of CVDs and divide them into hemorrhagic and ischemic conditions. Hemorrhagic diseases include arteriovenous malformations (AVM), intracranial aneurysms (IA), and intracranial hemorrhage (IH), whereas ischemic CVDs comprise atherosclerosis (AS), large vessel occlusion (LVO) / acute ischemic stroke (AIS) and Moya-moya disease (MMD). Although such diseases have different pathophysiological mechanisms, treatments, and prognoses, they all share a common characteristic: the diagnosis can be confirmed mainly using angiographic methods, whether CT angiography, MR angiography, or digital subtraction angiography (DSA).

The main objective of most works that use medical imaging is to extract semi-automatically or automatically features from an imaging modality that will be used in exploiting complex visual patterns associated with clinical outcomes and the classification at a study-, image- or pixel-level for a proper diagnosis. For this, rule-based algorithms or classification-based algorithms are used, the latter ranging from conventional machine learning ML algorithms to emerging deep learning DL algorithms. Although rule-based algorithm, such as Computer-Assisted Diagnostics (CAD) reflects a large portion of the various facets of AI for medical imaging, such tools have important constraints, such as the need for a detailed description of a set of rules detected by the human eye. An example is the identification of acute hemorrhage based on the establishment of an attenuation threshold above which pixels/voxels are selected. On the contrary, classification-based algorithms have the great advantage of identifying relevant features without requiring prior assumptions regarding their importance nor having to directly detail them, throughout the optimization of the error between the predicted and actual classification.⁶

Although ML techniques represented a significant advancement over CAD, it was still challenging to accurately model complex imaging data due to the rigid analytic forms of traditional ML algorithms. In light of that, less than a decade ago, advances in computer hardware enabled DL, a class of highly adaptable algorithms. DL has emerged as the state-of-the-art ML method with specific characteristics, such as much greater potential for complexity and capacity to theoretically model any arbitrary mathematical function, that reflects greater generalization capacity and scalability, as well as superior performance.

An overview of the ML algorithms used in medical data analysis is given in (Figure 1). Even though in the context of CVDs, conventional ML algorithms still seem to be more popular, as we move away from structured data - like that in electronic health records - to unstructured data -like that in medical imaging – DL potential to automatically extract underlying patterns directly from input images to predict target labels (without the need for feature vectors) gains relevance.⁶ Regardless of the clinical claim, it is clear that DL solutions—whether Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN)—are also heavily used in CVDs. On the one hand, the capability of CNNs to accept 2D or 3D images as input, in place of feature vectors, is its most novel feature. On the other hand, RNN and newer variations such as the long short-term memory network (LSTM) have achieved exceptional results for tasks requiring sequence labeling, in which time series/image order is important for the detection of an abnormality. Examples of such tasks in CVDs include diagnosis of diseases using MRI data composed of several images, consecutive frames in an ultrasound study (US), or multiple DSA frames with dynamic flow changes between them.⁷

Figure 1
Framework of common machine learning algorithms.

In the field of CVDs, AI approaches are primarily divided into detection, prediction (which mainly includes risk assessment and outcome prediction), and treatment assistance (Figure 2). Detection tasks include the location and segmentation of aneurysms, or identification and volume estimation of subtypes of acute hemorrhage.⁵,⁸ Prediction tasks generally include risk prediction and outcome estimation. An example of the former is the identification of high-risk patients for cerebral or retinal ischemic events based on clinical characteristics and ultrasonic plaque features.⁷ Examples of the latter are the prediction of the number/volume of lesions on DWI and T2/FLAIR days after an ischemic event based on the acute images, and prediction of functional independence (modified Rankin Scale ≤2) and good reperfusion (post-mTICI ≥2b) at 3 months after stroke using baseline clinical variables associated with treatment variables.⁹,¹⁰ For risk and outcome prediction tasks to perform well, as opposed to detection (which typically rely on a single modality as input), a variety of data must be integrated as input, including imaging findings with clinical, demographic, morphological, or hemodynamic characteristics. This favors the use of more complex models and architectures comprised of multiple integrated algorithms, that is, fusion models and multimodal AI. Finally, examples of treatment-aid tasks include real-time catheter monitoring and segmentation to facilitate greatly the delivery of guidewires later on or image fusion to more effectively direct bypass operations.⁵

Figure 2
Artificial Intelligence based applications in cerebrovascular diseases.

Hemorrhagic cerebrovascular diseases

Besides IH itself, the two hemorrhagic CVDs that have likely benefited the most from AI are aneurysm and AVM. Identifying these diseases using imaging scans is crucial for patient management since 10-20% of nontraumatic subarachnoid hemorrhage (SAH) are fatal and mainly caused by aneurysm and AVM rupture.¹¹

Concerning aneurysms, up to 2015, the great majority of papers for automatic detection used ruled-based or ML algorithms such as support vector machine (SVM), logistic regression (LR), Naïve Bayes, or random forest (RF).¹²,¹³ From 2017 on, studies focused on the use of DL models such as Res-Net, VGG, and U-Nets to identify and segment aneurysms, rather than in CTA, TOF MRA images, 2D-DSA or 3D rotational DSA (3DRA).¹⁴-¹⁷ In addition to the detection and segmentation, other classifiers combined image data with clinical, morphological, and hemodynamic information (e.g. wall shear stress, oscillatory shear) to predict individual rupture status of unruptured intracranial aneurysms and clinical outcome 6 months after flow diverted treatment.¹⁸-²⁰ CNNs have also been used to distinguish aneurysmal subarachnoid hemorrhages from other intracranial hemorrhage subtypes. Finally, ML techniques have been adopted to predict the outcome of aneurysmal SAH and the outcome after the rupture of the anterior communicating artery aneurysm.²¹,²²

Compared with aneurysms, very few studies on AVM have been published, although AVM nidus are not hard to detect. A number of studies combine various imaging approaches and modalities for AVM segmentation and assessment. They vary from center-line extraction methods (such as skeletonization approach) to unsupervised ML classifiers applied on CTA images and DL models applied on MRI images in order to differentiate, the nidus from brain tissue, and avoid unnecessary exposure of the healthy parenchyma to radiation during radiosurgery.²³-²⁵ In an interesting study, a system was built to assist AVM diagnosis.²⁶ For that, a faster RCNN was trained on 2D-DSA videos to track the appearance order of important vascular structures from the videos and quantify them as temporal features. Then, an SVM classifier combined this information with morphological features (obtained from radiomic methods applied in key frames) for AVM diagnosis and grading system.

Although 2D-DSA is still the "gold standard" for the diagnosis of AVM, new techniques, such as 3DRA and 4D-CTA have demonstrated similar value for AVM detection, identification of the main feeding and drain vessels, nidus delineation, and size estimation.²⁷,²⁸ Following Garcia et al.'s results, these modalities may be more thoroughly studied in AI models for AVM segmentation in the upcoming years.²⁹

Diverging a bit from morphological features, a study evidenced the robustness of quantitative DSA (QDSA) features of a selected ROI (e.g. peak density and time to peak) to predict the rupture of the AVM nidus.³⁰ For further prediction tasks, such as the prediction of post-operative results, the usage of other traditional methods is also discussed.³¹

Finally, computer systems to assist treatment in CVDs have been researched far less than for detection and prediction, since they usually require hardware implementation. Examples of these were introduced in a study to segment and track catheters on 2D X-ray sequences during endovascular aneurysm repair, and in other studies that introduced some novel conceptions of treatment in hemorrhagic CVDs, including automatic selection of interventional materials and real-time guidance in surgery.⁵,³²

Ischemic cerebrovascular diseases

Detection of large vessel occlusion (LVO) and AIS are two of the most pursuable objectives in applying AI to cerebrovascular pathology. Over the past decade, 13 companies have launched automated and semi-automated commercially available software for acute stroke and LVO diagnostics (Aidoc®, Apollo Medical Imaging Technology®, Brainomix®, inferVISION®, RAPID®, JLK Inspection®, Max-Q AI®, Nico.lab®, Olea Medical®, Qure.ai®, Viz.ai®, and Zebra Medical Vision®).³³ Also, considerable research work has been reported with the aid of ML models, mostly using medical images as input. While CT images have been used mainly for hemorrhage identification, LVO detection, and automated ASPECT calculation, MRI images have been used primarily for automatic core volume estimation on DWI images, evaluation of penumbra, and to predict the final ischemic stroke lesions from initial MRI, stroke symptom onset and hemorrhagic transformation.³⁴-⁴¹ AI has also been used to improve image quality and speed acquisition - since time delays associated with brain scanning is a real constrain – for stroke risk prediction, for analysis of cerebral edema, and to evaluate treatment effect.⁴²-⁴⁴

Except for stroke, to the best of our knowledge, a small number of essays discuss the computer-assisted analysis of other ischemic CVDs such AS and MMD, and for these abnormalities, models tend not to use medical pictures as input. This might be because the diagnostic criteria for AS, for instance, are much more complex and less standardized than the morphological or gray-level attributes used to identify aneurysms. Examples are ML algorithms trained to detect the degree of carotid atherosclerosis (CAS) using clinical indicators, including age, gender, hypertension, cholesterol, and glucose levels, as well as pulse wave features.⁴⁵,⁴⁶ Similarly, ML and DL were used in a dataset of 636 subjects, to diagnose AS while relying solely on a few clinical criteria and reaching an accuracy of 96.4%.⁴⁷

Regarding carotid plaque tissue characterization, a systematic review evidenced that US imaging was the most commonly used imaging technique for applying AI methods (94.59%), followed by CT (30%) and MRI (35%).⁷ Most articles used manual and statistical analysis, followed by ML (especially SVM, RF, and DT). In addition, the most common architecture used in the DL framework for all three imaging modalities was U-Net. Although DL is a modality that is still emerging, it might be promising to address some of the challenges of plaque evaluation, since it can characterize plaque's composition in spite of variations and fuzziness in pixel distribution to be non-linear. In fact, a CNN was proposed to classify plaque components into lipid core, fibrous cap, and calcified tissue and achieved a correlation coefficient of 0.90 between automatic measurement and expert measurement.⁴⁸ Finally, since CVDs cover both stroke and myocardial infarction, carotid intraplaque neovascularization in B-mode ultrasonography have been applied to multiclass ML to predict coronary artery disease (CAD).⁴⁹ Such an automated system is likely to become a prominent CAD detection system in the future.

Lastly, various medical imaging modalities have been used in pertinent DL-based applications for MMD automatic detection, including plain skull radiography, T2-weighted MRI and DSA images.⁵⁰-⁵² DL and sparse representation-based classification were also used to predict the hemorrhagic risk of MMD in DSA and to assess vascular cognitive impairment on resting-state functional MRI, respectively.⁵³,⁵⁴

AI IN EPILEPSY

Epilepsy affects nearly 70 million people around the world, with a huge social and economic burden.⁵⁵ Adequate epilepsy care usually requires a multidisciplinary work-up team that seeks a precise diagnosis, comprehension of cognitive comorbidities and impact on quality of life, response to pharmacological treatment, and decisions regarding surgical management and prognosis. It is also important to bear in mind that epilepsy may be caused by distinct underlying diseases, such as hippocampal sclerosis, malformations of cortical development, infectious diseases, autoimmune conditions, brain scars, and tumors.⁵⁶ While positron emission tomography (PET), single photon emission computed tomography (SPECT), and magnetoencephalography may help to localize cerebral dysfunction, magnetic resonance imaging (MRI) may identify and accurately delineate underlying specific pathologies, some of them amenable to surgical treatment. However, many unanswered questions remain in the debate of epilepsy care, such as cryptogenic seizures, the involvement of otherwise apparently normal brain parenchyma, and complex networks between epileptogenic foci and diverse cognitive deficits.

A combination of statistics, data science, and computational resources may yield big-data problem-solving and shed light on a more comprehensive approach to several neurological conditions. Therefore, a collaboration of scientists in multicentric studies has generated advancements in global neuroscience, such as the Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA) Consortium.⁵⁶ The ENIGMA-Epilepsy working group has applied advanced methods such as structural covariance and event-based-modeling, combined with DL approaches to analyze large datasets of structural MRI, diffusion tensor imaging (DTI), and resting state functional MRI of patients with epilepsy and healthy controls.⁵⁷ This group recently demonstrated that multivariate ML approaches could successfully classify healthy controls and patients with temporal lobe epilepsy (TLE) secondary to hippocampal sclerosis (HS), as well as to lateralize the side of TLE, using morphological features from T1-weighted data and diffusion MRI parameters.⁵⁸

Another intriguing study showed that NeuroQuant, an FDA-approved automated software, had lower sensitivity and negative predictive values but had specificity and positive predictive values comparable to neuroradiologists' visual assessments in the detection of HS in a large cohort of TLE patients undergoing presurgical evaluation.⁵⁹ Although studies so far do not demonstrate that quantitative imaging can replace visual MRI assessment by experienced neuroradiologists in the detection of HS, there may be a role for the use of these tools in the preliminary evaluation of surgical candidates in centers that lack such professionals.

Malformations of cortical development (MCD) account for the most common surgical pathologies in children and the third most common cause in adults with pharmacoresistant focal epilepsies.⁶⁰ Out of them, focal cortical dysplasia represents the vast majority, and its diagnosis heavily depends on both the skill of the examiners and the optimization of MRI protocols, which are preferentially acquired using high-field strength scanners.⁶¹ Post-processing tools show promise for assisting in the detection of more subtle lesions, but some cases still remain overlooked.⁶² Recently, one multicentric study validated the use of a DL algorithm to detect MRI-negative FCD with multimodal MRI data.⁶³

It is well known that seizures are not the sole component of epilepsy that affects the quality of life of patients and their families. Affected individuals may also suffer from deficits in several cognitive domains, including language and memory, and some studies using ML algorithms have accurately predicted cognitive comorbidities.⁶⁴,⁶⁵ Interestingly, AI has been used to predict brain age in patients with epilepsy. These studies revealed that patients had significantly older brain ages than healthy controls, and functional brain age was related to decreased cognition capabilities.⁶⁶,⁶⁷

It seems that ML and DL algorithms might also be useful in the prediction of clinical prognosis after surgical treatment of medically refractory epilepsy. Previous works that incorporated quantitative volumetric MRI measurements demonstrated that subtle cortical brain atrophy beyond the surgical site influences seizure-free outcomes, reinforcing the idea that focal epilepsies are complex network disorders.⁶⁸,⁶⁹ In addition, DL methods were applied to structural connectomes to predict postsurgical seizure outcomes at least one year after epilepsy surgery.⁷⁰

Despite the emerging AI tools in epilepsy, future works are needed to clarify if these applications are not only a transient hype and if they will ultimately prove to be useful in care at an individual end-user level.

AI IN MULTIPLE SCLEROSIS

MRI has a crucial role in the diagnosis and monitoring of multiple sclerosis (MS) patients. The definition of MS according to the McDonald criteria requires the clinical and/or MRI identification of central nervous system (CNS) lesions disseminated in time (DIT) and space (DIS).⁷¹ However, with the recent definition of "new" demyelinating entities, such as neuromyelitis optica spectrum disorders (NMOSD) and myelin-oligodendrocyte glycoprotein antibody-associated disease (MOGAD), there is an increasing need for a more precise description of lesion shapes and the use of new MRI biomarkers with a better representation of the pathological substrate of MS, such as central vein sign, cortical lesions and paramagnetic rim lesions.⁷²-⁷⁵

The primary goal in the follow-up and control of MS patients is to achieve the status of no evidence of disease activity (NEDA), which essentially consists of the absence of clinical symptoms or progression, new or expanding T2-FLAIR demyelinating lesions, and no new T1-gadolinium enhanced lesions.⁷⁶ This highlights the need for a precise lesion count in MRI readings. Atrophy measurements are also becoming increasingly important in the MRI follow-up of MS patients as the relevance of the degenerative processes for the course of the disease is recognized.⁷⁶ However, in everyday practice, several of these measurements, including lesion count and analysis of changes in lesion volumes, might be very time-consuming. Manual measurements of the brain and spinal cord volume can take even more time and might also be imprecise and with low reproducibility. Artificial intelligence techniques could therefore be helpful for a faster and more precise quantification and follow-up of MS patients.

In the last few years, the use of AI in MS to foresee diagnosis, forecast long-term prognosis and ensure trustworthy findings and time efficiency has received a lot of attention in recent years. The use of AI in medical imaging, particularly MRI, has shown encouraging results, enabling automated lesion and tissue segmentation, disease categorization, and contract synthetization from advanced sequences. Such a strategy is also appropriate for the emerging field of "omics," where the evaluation of enormous data sets gathered from a single patient is essential from the standpoint of personalized medicine.⁷⁷

AI can be used to interpret and analyze MR images of patients with MS, detecting subtle changes in the images of central nervous system structures, including the brain and spinal cord, over time. ML algorithms can be trained on large MRI image datasets to recognize specific patterns and features associated with MS. These algorithms can identify brain and spinal cord lesions characteristic of MS, such as demyelinating plaques, with high precision and speed, as well as track the emergence of new lesions and the increase in lesion volume by analyzing time series of MR images.⁷⁸

New advances in the automation of combined methods, the traditional unsupervised machine learning technique, and the DL attention-gate 3D U-net network have improved the detection of MRI lesions in infratentorial and juxtacortical regions.⁷⁹ ML algorithms are trained using clinical, imaging, and genetic data to identify patterns and characteristics that are indicative of disease in order to make predictions or decisions. DL can be applied to analyze MR images at a deeper level, identifying subtle and complex features of the MS-related lesions and MRI normal-appearing tissues. A great advantage of DL is the ability to handle large amounts of complex, unstructured data, such as high-resolution medical images. However, training deep neural networks usually requires a large set of annotated data and considerable computational resources.⁸⁰,⁸¹

The central vein sign (CVS) has emerged as a promising MRI biomarker to improve the accuracy and speed the diagnosis of MS. The introduction of the CVS concept has added a new dimension to the diagnostic capabilities of MRI in MS, distinguishing MS lesions from non-MS white matter lesions that mimic its clinical and radiological features. The presence of a central vein within an MS lesion is thought to reflect the underlying pathological process of MS, which involves inflammation and destruction of the myelin surrounding the veins.⁷⁴ The veins appear as dark signal voids on T2-weighted or T2*-weighted MRI sequences due to the flow-void effect.⁸² The CVSnet, a DL-based prototype for automated evaluation of the CVS in white matter MS lesions, was tested using data from several hospitals and has since allowed for bigger multicenter trials to determine the merit of including the CVS marker in MS diagnostic criteria.⁸³

There are several commercial volumetric MRI analysis solutions that can assist in lesion identification and segmentation by providing automatized, quantitative, and qualitative measurements for the diagnosis and monitoring of MS. Among others, CorTechs.ai, Icometrix, Qynapse, Pixyl are examples of such tools that received regulatory approval and have helped with clinical decision-making and the development of personalized therapies.⁸⁴

Another promising application of AI in MS diagnosis is the analysis of data from sensors and wearable devices. These devices can collect a variety of data such as movement, heart rate, and sleep quality. AI algorithms can process this data and identify patterns that could be related to MS symptoms such as fatigue, balance issues, and gait changes. This information can be useful both for the diagnosis and ongoing monitoring of patients with MS.⁸⁰,⁸⁵

Although some of these tools are validated in compelling databases, it is still important to understand their impact in the clinical setting in different populations. Furthermore, it is intuitive that a faster and more precise diagnostic might be provided by AI, supposedly having a positive economic impact on healthcare.⁸⁶ However, the use of AI also implies additional costs and more and wider studies of the economic impact of the use of AI in MS are needed.

In summary, AI offers useful data that speed up the identification and measurement of targeted lesions as well as the measurement of atrophy and degenerative processes. It is crucial to emphasize though, that AI is only a supporting tool; medical professionals are still ultimately responsible for the diagnosis, and the final interpretation of MRI scans. Clinical experience and medical judgment are still essential for medical care and to guide the future of AI.⁸⁷

AI IN NEURODEGENERATIVE DISEASES

Neurodegenerative diseases involve different clinical conditions with various pathological substrates that, in general, result in significant disability with major social and economic impact. The diverse group of neurodegenerative diseases, which have shown an increase in prevalence with population aging, includes dementia syndromes such as Alzheimer's disease (AD), movement disorders like Parkinson's disease (PD) and atypical parkinsonism, as well as Amyotrophic Lateral Sclerosis (ALS).⁸⁸

Since the clinical manifestations of many of these disorders overlap and a firm anatomopathological diagnosis is impractical, diagnosing them might be difficult. It is not uncommon for the proper diagnosis to be delayed, which affects the prognosis. In such instances, neuroimaging has gained significant relevance, particularly structural MRI and functional Nuclear Medicine.⁸⁹

The main role of MRI is to exclude other causes that might justify the symptoms, such as sequelae of vascular/traumatic events, tumors, hydrocephalus, among others.⁹⁰ Moreover, the development of MRI sequences and high-field MRI scanners like the 3T has led to novel imaging discoveries that have been supportive in either ruling out or confirming clinical hypotheses.

In addition, AI advances, whether through quantification methods or DL techniques, will be essential for assessing neurodegenerative diseases in a number of ways, including detecting the onset of the disease, characterizing it, aiding in differential diagnoses, quantifying disease progression, and evaluating treatment response.

Dementia syndromes

Evaluation of dementia syndromes by structural magnetic resonance allows for identifying atrophy patterns that predict assertive diagnosis. Volumetric T1-weighted sequence plays a fundamental role, providing adequate spatial resolution and cortical atrophy pattern characterization. However, due to the lack of specific findings, specific diagnosis of some dementia faces some resistance, and incipient imaging findings are frequently neglected.⁹⁰ In this context, it was mandatory to develop systematic and practical analysis, with assessment scales, aiming for interpersonal uniformity. For example, the evaluation of medial temporal atrophy in AD through Scheltens scale and, more recently, through the visual scale described by Urs et al.⁹¹

Visual scales limitations are not neglectable, and automatic measurement and stratification systems have been improved. Although some platforms are expensive to use, they can deliver rapid results. Through quantification techniques, supported by AI and the OASIS database, some studies have shown up to 96.85% accuracy in detecting AD compared to healthy controls, 84.3% in differentiating between AD and frontotemporal dementia, and 97.48% in differentiating between AD and mild cognitive decline.⁹²

Parkinson's disease and other movement disorders

The role of MRI has grown in importance over the past few years in the setting of PD and its differential diagnosis, particularly Progressive Supranuclear Palsy (PSP) and Multiple Systems Atrophy (MSA), despite not being included in the Movement Disorder Society diagnostic criteria.⁹³

With the advent of the 3T and high-resolution sequences, in 2014 the loss of the swallow tail sign was depicted, with high specificity and sensitivity for the identification of parkinsonism.⁹⁴ It reflects the loss of susceptibility sequence hypersignal in dorsolateral substantia nigra pars compacta (nigrosome-1 topography) due to probable iron deposits and neuromelanin reduction. Despite being considered a biomarker of PD, the loss of the swallow sign is not a diagnostic criterion, as it is found in atypical parkinsonism. However, the negative predictive value of 100% suggests that the presence of the sign makes the diagnosis of PD much less likely.

Furthermore, the evaluation of neuromelanin in the substantia nigra, through magnetization transfer T1-weighted sequences (T1-MTC), is also considered a biomarker of PD, useful even in the evaluation of progression.⁹⁵ A 2022 study demonstrated that neuromelanin quantification is progressively reduced among patients with isolated REM sleep behavior disorder (considered a prodromal stage of PD) and patients with established PD compared to healthy controls.⁴⁷ In addition, DL was used to delineate automatically the regions of interest in the substantia nigra pars compacta (SNc) and demonstrated comparable results to manual measurement, but with the advantage of being faster.⁹⁶

Despite advances in AI research in PD, its use is still limited due to several constraints, including small samples, low accuracy, and limitations specific to the ML algorithms.⁹⁷ MRI also allows the evaluation of other PSP and MSA imaging findings, such as brainstem morphometric rates and depiction of pontine hot cross bun sign and putaminal slit sign.⁹³

Finally, studies using a support vector machine (SVM) demonstrated accuracy, specificity, and sensitivity of 91%, 88%, and 93%, respectively, in distinguishing MSA and PD patients. Another study that similarly used SVM models achieved an accuracy of 91.7% for MSA detection using mesencephalic evaluation.⁹²

Amyotrophic lateral sclerosis

ALS is a neurodegenerative disease that affects upper and lower motor neurons. Its prognosis is quite limited, with life expectancy ranging between 3 to 5 years after the onset of symptoms.⁹⁸ The diagnosis of lower motor neuron involvement is based on electroneuromyography and muscle biopsy. In contrast, the diagnosis of upper motor neuron involvement is challenging and the role of MRI must be highlighted. In 2012, T1-MTC hyperintensity along the corticospinal tract (CST) was described, demonstrating high specificity (100%), despite low sensitivity (37%).⁹⁹

Advances in AI in ALS have so far focused on the diagnosis, mainly on the differential diagnosis with other neurodegenerative diseases that also affect motor function, such as PD, Huntington's disease, and MSA. Also, AI has been useful in understanding disease pathological processes. A recent study demonstrated an accuracy of 70-94% for distinguishing four clinical/radiological phenotypes of ALS patients (with signal alteration on CTS, without signal alteration on CST, classic patients with obvious lower motor neuron and upper motor neuron clinical signs, and patients with frontotemporal dementia) using a Random Forest (RF) algorithm.¹⁰⁰ Another study also showed an accuracy of 78.7% in the estimation of patient survival, mainly in the short survival group, through the use of MR sequences (DTI combined with T1-weighted images) and DL.⁹⁸

What we recognize is that, in addition to the advancement of neuroimaging, especially with regard to technical sequences evolution, analysis, and report standardization, the complete evaluation of neurodegenerative patients is essential for diagnosis or follow-up. Future projections in the set of neurodegenerative diseases management aim at earlier diagnosis, in prodromal phases, so that specific therapies, still under development, may impact on disease's natural history, before neuronal loss and consolidated disability. Additionally, it has been argued that AI may help in providing precision, agility, and even solutions to a number of unresolved questions.

In conclusion, AI exhibits significant potential for advancing the field of neuroimaging, offering transformative solutions for both diagnostic and therapeutic applications. Despite this promise, the integration of AI algorithms into clinical practice is currently confined to a limited number of specialized research institutions. This limitation is principally attributed to two critical barriers:

• the lack of algorithmic generalizability across diverse datasets and;

• the absence of sustainable business models for widespread adoption.

As technological innovations continue to address these challenges, it is anticipated that an increasing number of AI-based models will transition from experimental stages to clinical implementation. Another crucial aspect of AI in medicine is that even an exemplary AI model is not a guarantee for improved clinical care outcomes. Both under-reliance and over-reliance on AI algorithms can compromise their efficacy, either by failing to capitalize on their diagnostic or therapeutic capabilities, or by diminishing the role of human clinical expertise in medical decision-making, respectively. The way clinicians engage with AI algorithms critically influences the potential for either substantial benefits or detriments in healthcare outcomes.

References

• ¹
  Lui YW, Chang PD, Zaharchuk G, et al. Artificial Intelligence in Neuroradiology: Current Status and Future Directions. AJNR Am J Neuroradiol 2020;41(08):E52–E59. Doi: 10.3174/ajnr.A6681
  » https://doi.org/10.3174/ajnr.A6681
• ²
  Fiani B, Pasko KBD, Sarhadi K, Covarrubias C. Current uses, emerging applications, and clinical integration of artificial intelligence in neuroradiology. Rev Neurosci 2021;33(04):383–395. Doi: 10.1515/revneuro-2021-0101
  » https://doi.org/10.1515/revneuro-2021-0101
• ³
  Duong MT, Rauschecker AM, Mohan S. Diverse Applications of Artificial Intelligence in Neuroradiology. Neuroimaging Clin N Am 2020;30(04):505–516. Doi: 10.1016/j.nic.2020.07.003
  » https://doi.org/10.1016/j.nic.2020.07.003
• ⁴
  Olthof AW, van Ooijen PMA, Rezazade Mehrizi MH. Promises of artificial intelligence in neuroradiology: a systematic technographic review. Neuroradiology 2020;62(10):1265–1278. Doi: 10.1007/s00234-020-02424-w
  » https://doi.org/10.1007/s00234-020-02424-w
• ⁵
  Chen X, Lei Y, Su J, et al. A Review of Artificial Intelligence in Cerebrovascular Disease Imaging: Applications and Challenges. Curr Neuropharmacol 2022;20(07):1359–1382. Doi: 10.2174/1570159X19666211108141446
  » https://doi.org/10.2174/1570159X19666211108141446
• ⁶
  Segato A, Marzullo A, Calimeri F, De Momi E. Artificial intelligence for brain diseases: A systematic review. APL Bioeng 2020;4 (04):041503. Doi: 10.1063/5.0011697
  » https://doi.org/10.1063/5.0011697
• ⁷
  Saba L, Sanagala SS, Gupta SK, et al. Multimodality carotid plaque tissue characterization and classification in the artificial intelligence paradigm: a narrative review for stroke application. Ann Transl Med 2021;9(14):1206. Doi: 10.21037/atm-20-7676
  » https://doi.org/10.21037/atm-20-7676
• ⁸
  Monteiro M, Newcombe VFJ, Mathieu F, et al. Multiclass semantic segmentation and quantification of traumatic brain injury lesions on head CT using deep learning: an algorithm development and multicentre validation study. Lancet Digit Health 2020;2(06):e314–e322. Doi: 10.1016/S2589-7500(20)30085-6
  » https://doi.org/10.1016/S2589-7500(20)30085-6
• ⁹
  Mouridsen K, Thurner P, Zaharchuk G. Artificial Intelligence Applications in Stroke. Stroke 2020;51(08):2573–2579. Doi: 10.1161/STROKEAHA.119.027479
  » https://doi.org/10.1161/STROKEAHA.119.027479
• ¹⁰
  van Os HJA, Ramos LA, Hilbert A, et al; MR CLEAN Registry Investigators. Predicting Outcome of Endovascular Treatment for Acute Ischemic Stroke: Potential Value of Machine Learning Algorithms. Front Neurol 2018;9:784. Doi: 10.3389/fneur.2018.00784
  » https://doi.org/10.3389/fneur.2018.00784
• ¹¹
  Suarez JI, Tarr RW, Selman WR. Aneurysmal subarachnoid hemorrhage. N Engl J Med 2006;354(04):387–396. Doi: 10.1056/NEJMra052732
  » https://doi.org/10.1056/NEJMra052732
• ¹²
  Yang X, Blezek DJ, Cheng LT, Ryan WJ, Kallmes DF, Erickson BJ. Computer-aided detection of intracranial aneurysms in MR angiography. J Digit Imaging 2011;24(01):86–95. Doi: 10.1007/s10278-009-9254-0
  » https://doi.org/10.1007/s10278-009-9254-0
• ¹³
  Malik KM, Anjum SM, Soltanian-Zadeh H, Malik H, Malik GM. A Framework for Intracranial Saccular Aneurysm Detection and Quantification using Morphological Analysis of Cerebral Angiograms. IEEE Access 2018;6:7970–7986. Doi: 10.1109/ACCESS.2018.2799307
  » https://doi.org/10.1109/ACCESS.2018.2799307
• ¹⁴
  Shi Z, Miao C, Schoepf UJ, et al. A clinically applicable deep-learning model for detecting intracranial aneurysm in computed tomography angiography images. Nat Commun 2020;11(01): 6090. Doi: 10.1038/s41467-020-19527-w
  » https://doi.org/10.1038/s41467-020-19527-w
• ¹⁵
  Ueda D, Yamamoto A, Nishimori M, et al. Deep Learning for MR Angiography: Automated Detection of Cerebral Aneurysms. Radiology 2019;290(01):187–194. Doi: 10.1148/radiol.2018180901
  » https://doi.org/10.1148/radiol.2018180901
• ¹⁶
  Jin H, Geng J, Yin Y, et al. Fully automated intracranial aneurysm detection and segmentation from digital subtraction angiography series using an end-to-end spatiotemporal deep neural network. J Neurointerv Surg 2020;12(10):1023–1027. Doi: 10.1136/neurintsurg-2020-015824
  » https://doi.org/10.1136/neurintsurg-2020-015824
• ¹⁷
  Zeng Y, Liu X, Xiao N, et al. Automatic Diagnosis Based on Spatial Information Fusion Feature for Intracranial Aneurysm. IEEE Trans Med Imaging 2020;39(05):1448–1458. Doi: 10.1109/ TMI.2019.2951439
  » https://doi.org/10.1109/TMI.2019.2951439
• ¹⁸
  Chen G, Lu M, Shi Z, et al. Development and validation of machine learning prediction model based on computed tomography angiography-derived hemodynamics for rupture status of intracranial aneurysms: a Chinese multicenter study. Eur Radiol 2020;30(09):5170–5182. Doi: 10.1007/s00330-020-06886-7
  » https://doi.org/10.1007/s00330-020-06886-7
• ¹⁹
  Paliwal N, Jaiswal P, Tutino VM, et al. Outcome prediction of intracranial aneurysm treatment by flow diverters using machine learning. Neurosurg Focus 2018;45(05):E7. Doi: 10.3171/2018.8. FOCUS18332
  » https://doi.org/10.3171/2018.8.FOCUS18332
• ²⁰
  Cho J, Park KS, Karki M, et al. Improving Sensitivity on Identification and Delineation of Intracranial Hemorrhage Lesion Using Cascaded Deep Learning Models. J Digit Imaging 2019;32(03): 450–461. Doi: 10.1007/s10278-018-00172-1
  » https://doi.org/10.1007/s10278-018-00172-1
• ²¹
  de Toledo P, Rios PM, Ledezma A, Sanchis A, Alen JF, Lagares A. Predicting the outcome of patients with subarachnoid hemorrhage using machine learning techniques. IEEE Trans Inf Technol Biomed 2009;13(05):794–801. Doi: 10.1109/TITB.2009.2020434
  » https://doi.org/10.1109/TITB.2009.2020434
• ²²
  Xia N, Chen J, Zhan C, et al. Prediction of Clinical Outcome at Discharge After Rupture of Anterior Communicating Artery Aneurysm Using the Random Forest Technique. Front Neurol 2020;11:538052. Doi: 10.3389/fneur.2020.538052
  » https://doi.org/10.3389/fneur.2020.538052
• ²³
  Babin D, Spyrantis M, Pizurica A, Philips W Pixel profiling for extraction of arteriovenous malformation in 3-D CTA images. Conference proceedings: Conference proceedings : Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Conference. 2013. 5449-5452 Doi: 10.1109/EMBC.2013.6610782
  » https://doi.org/10.1109/EMBC.2013.6610782
• ²⁴
  Peng SJ, Lee CC, Wu HM, et al. Fully automated tissue segmentation of the prescription isodose region delineated through the Gamma knife plan for cerebral arteriovenous malformation (AVM) using fuzzy C-means (FCM) clustering. Neuroimage Clin 2019;21:101608. Doi: 10.1016/j.nicl.2018.11.018
  » https://doi.org/10.1016/j.nicl.2018.11.018
• ²⁵
  Simon AB, Hurt B, Karunamuni R, et al. Automated segmentation of multiparametric magnetic resonance images for cerebral AVM radiosurgery planning: a deep learning approach. Sci Rep 2022; 12(01):786. Doi: 10.1038/s41598-021-04466-3
  » https://doi.org/10.1038/s41598-021-04466-3
• ²⁶
  Shi K, Xiao W, Wu G, et al. Temporal-Spatial Feature Extraction of DSA Video and Its Application in AVM Diagnosis. Front Neurol 2021;12:655523. Doi: 10.3389/fneur.2021.655523
  » https://doi.org/10.3389/fneur.2021.655523
• ²⁷
  Wang H, Ye X, Gao X, Zhou S, Lin Z. The diagnosis of arteriovenous malformations by 4D-CTA: a clinical study. J Neuroradiol 2014; 41(02):117–123. Doi: 10.1016/j.neurad.2013.04.004
  » https://doi.org/10.1016/j.neurad.2013.04.004
• ²⁸
  Anderson JL, Khattab MH, Sherry AD, et al. Improved Cerebral Arteriovenous Malformation Obliteration With 3-Dimensional Rotational Digital Subtraction Angiography for Radiosurgical Planning: A Retrospective Cohort Study. Neurosurgery 2020; 88(01):122–130. Doi: 10.1093/neuros/nyaa321
  » https://doi.org/10.1093/neuros/nyaa321
• ²⁹
  Garcia C, Fang Y-B, Liu J, Narata A, Orlando J, Larrabide I. A deep learning model for brain vessel segmentation in 3DRA with arteriovenous malformations. SPIE Digital Library. 2022;28;. Doi: 10.48550/arXiv.2210.02416
  » https://doi.org/10.48550/arXiv.2210.02416
• ³⁰
  Hong JS, Lin CJ, Lin YH, et al. Machine Learning Application With Quantitative Digital Subtraction Angiography for Detection of Hemorrhagic Brain Arteriovenous Malformations. IEEE Access 2020;8:204573–204584. Doi: 10.1109/ACCESS.2020.3036692
  » https://doi.org/10.1109/ACCESS.2020.3036692
• ³¹
  Oermann EK, Rubinsteyn A, Ding D, et al. Using a Machine Learning Approach to Predict Outcomes after Radiosurgery for Cerebral Arteriovenous Malformations. Sci Rep 2016;6:21161. Doi: 10.1038/srep21161
  » https://doi.org/10.1038/srep21161
• ³²
  Zhou Y-J, Xie X-L, Hou Z-G, Bian G-B, Liu S-Q, Zhou X-H FRR-NET: Fast Recurrent Residual Networks for Real-Time Catheter Segmentation and Tracking in Endovascular Aneurysm Repair," In: 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI); 2020; Iowa City, USA. pp. 961-964 Doi: 10.1109/ ISBI45749.2020.9098632
  » https://doi.org/10.1109/ISBI45749.2020.9098632
• ³³
  Mainali S, Darsie ME, Smetana KS. Machine Learning in Action: Stroke Diagnosis and Outcome Prediction. Front Neurol 2021; 12:734345. Doi: 10.3389/fneur.2021.734345
  » https://doi.org/10.3389/fneur.2021.734345
• ³⁴
  Kuo W, Hӓne C, Mukherjee P, Malik J, Yuh EL. Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning. Proc Natl Acad Sci U S A 2019; 116(45):22737–22745
• ³⁵
  Chen Z, Zhang R, Xu F, et al. Novel Prehospital Prediction Model of Large Vessel Occlusion Using Artificial Neural Network. Front Aging Neurosci 2018;10:181. Doi: 10.3389/fnagi.2018.00181, Erratum in: Front Aging Neurosci. 2018 Jul 17;10:222. PMID: 29997494; PMCID: PMC6028566
  » https://doi.org/10.3389/fnagi.2018.00181
• ³⁶
  Nagel S, Joly O, Pfaff J, et al. e-ASPECTS derived acute ischemic volumes on non-contrast-enhanced computed tomography images. Int J Stroke 2020;15(09):995–1001. Doi: 10.1177/ 1747493019879661
  » https://doi.org/10.1177/1747493019879661
• ³⁷
  Bridge CP, Bizzo BC, Hillis JM, et al. Development and clinical application of a deep learning model to identify acute infarct on magnetic resonance imaging. Sci Rep 2022;12(01):2154. Doi: 10.1038/s41598-022-06021-0
  » https://doi.org/10.1038/s41598-022-06021-0
• ³⁸
  Chen L, Bentley P, Rueckert D. Fully automatic acute ischemic lesion segmentation in DWI using convolutional neural networks. Neuroimage Clin 2017;15:633–643. Doi: 10.1016/j. nicl.2017.06.016
  » https://doi.org/10.1016/j.nicl.2017.06.016
• ³⁹
  Yu Y, Xie Y, Thamm T, et al. Use of Deep Learning to Predict Final Ischemic Stroke Lesions From Initial Magnetic Resonance Imaging. JAMA Netw Open 2020;3(03):e200772. Doi: 10.1001/ jamanetworkopen.2020.0772, Erratum in: JAMA Netw Open. 2020 Oct 1;3(10):e2026464. PMID: 32163165; PMCID: PMC7068232
  » https://doi.org/10.1001/jamanetworkopen.2020.0772
• ⁴⁰
  Ho KC, Speier W, Zhang H, Scalzo F, El-Saden S, Arnold CW. A Machine Learning Approach for Classifying Ischemic Stroke Onset Time From Imaging. IEEE Trans Med Imaging 2019;38 (07):1666–1676. Doi: 10.1109/TMI.2019.2901445
  » https://doi.org/10.1109/TMI.2019.2901445
• ⁴¹
  Yu Y, Guo D, Lou M, Liebeskind D, Scalzo F. Prediction of Hemorrhagic Transformation Severity in Acute Stroke From Source Perfusion MRI. IEEE Trans Biomed Eng 2018;65(09): 2058–2065. Doi: 10.1109/TBME.2017.2783241
  » https://doi.org/10.1109/TBME.2017.2783241
• ⁴²
  Dritsas E, Trigka M. Stroke Risk Prediction with Machine Learning Techniques. Sensors (Basel) 2022;22(13):4670. Doi: 10.3390/ s22134670
  » https://doi.org/10.3390/s22134670
• ⁴³
  Dhar R, Chen Y, An H, Lee JM. Application of Machine Learning to Automated Analysis of Cerebral Edema in Large Cohorts of Ischemic Stroke Patients. Front Neurol 2018;9:687. Doi: 10.3389/fneur.2018.00687
  » https://doi.org/10.3389/fneur.2018.00687
• ⁴⁴
  Nielsen A, Hansen MB, Tietze A, Mouridsen K. Prediction of Tissue Outcome and Assessment of Treatment Effect in Acute Ischemic Stroke Using Deep Learning. Stroke 2018;49(06): 1394–1401. Doi: 10.1161/STROKEAHA.117.019740
  » https://doi.org/10.1161/STROKEAHA.117.019740
• ⁴⁵
  Yu J, Zhou Y, Yang Q, et al. Machine learning models for screening carotid atherosclerosis in asymptomatic adults. Sci Rep 2021;11 (01):22236. Doi: 10.1038/s41598-021-01456-3
  » https://doi.org/10.1038/s41598-021-01456-3
• ⁴⁶
  Ding L, Zhou R, Liu G Eds. Study on the classification algorithm of degree of arteriosclerosis based on fuzzy pattern recognition. In: International Conference on Image Processing and Pattern Recognition in Industrial Engineering; 2010; Xi’an, China. 78200B. https://doi.org/10.1117/12.867445
  » https://doi.org/10.1117/12.867445
• ⁴⁷
  Terrada O, Cherradi B, Raihani A, Bouattane O Atherosclerosis disease prediction using Supervised Machine Learning Techniques. In: 1st International Conference on Innovative Research in Applied Science, Engineering and Technology (IRASET); 2020; Meknes; Morocco. pp. 1-5 Doi: 10.1109/IRASET48871.2020.9092082
  » https://doi.org/10.1109/IRASET48871.2020.9092082
• ⁴⁸
  Lekadir K, Galimzianova A, Betriu A, et al. A Convolutional Neural Network for Automatic Characterization of Plaque Composition in Carotid Ultrasound. IEEE J Biomed Health Inform 2017;21(01): 48–55. Doi: 10.1109/JBHI.2016.2631401
  » https://doi.org/10.1109/JBHI.2016.2631401
• ⁴⁹
  Jamthikar AD, Gupta D, Mantella LE, et al. Multiclass machine learning vs. conventional calculators for stroke/CVD risk assessment using carotid plaque predictors with coronary angiography scores as gold standard: a 500 participants study. Int J Cardiovasc Imaging 2021;37(04):1171–1187. Doi: 10.1007/s10554-020-02099-7
  » https://doi.org/10.1007/s10554-020-02099-7
• ⁵⁰
  Kim T, Heo J, Jang DK, et al. Machine learning for detecting moyamoya disease in plain skull radiography using a convolutional neural network. EBioMedicine 2019;40:636–642. Doi: 10.1016/j.ebiom.2018.12.043
  » https://doi.org/10.1016/j.ebiom.2018.12.043
• ⁵¹
  Akiyama Y, Mikami T, Mikuni N. Deep Learning-Based Approach for the Diagnosis of Moyamoya Disease. J Stroke Cerebrovasc Dis 2020;29(12):105322. Doi: 10.1016/j.jstrokecerebrovasdis.2020. 105322
  » https://doi.org/10.1016/j.jstrokecerebrovasdis.2020.105322
• ⁵²
  Hu T, Lei Y, Su J, et al. Learning spatiotemporal features of DSA using 3D CNN and BiConvGRU for ischemic moyamoya disease detection. Int J Neurosci 2023;133(05):512–522. Doi: 10.1080/ 00207454.2021.1929214
  » https://doi.org/10.1080/00207454.2021.1929214
• ⁵³
  Lei Y, Zhang X, Ni W, et al. Recognition of moyamoya disease and its hemorrhagic risk using deep learning algorithms: sourced from retrospective studies. Neural Regen Res 2021;16(05): 830–835. Doi: 10.4103/1673-5374.297085
  » https://doi.org/10.4103/1673-5374.297085
• ⁵⁴
  Lei Y, Chen X, Su JB, et al. Recognition of Cognitive Impairment in Adult Moyamoya Disease: A Classifier Based on High-Order Resting-State Functional Connectivity Network. Front Neural Circuits 2020;14:603208. Doi: 10.3389/fncir.2020.603208
  » https://doi.org/10.3389/fncir.2020.603208
• ⁵⁵
  Thijs RD, Surges R, O’Brien TJ, Sander JW. Epilepsy in adults. Lancet 2019;393(10172):689–701. Doi: 10.1016/S0140-6736(18)32596-0
  » https://doi.org/10.1016/S0140-6736(18)32596-0
• ⁵⁶
  Thompson PM, Jahanshad N, Ching CRK, et al; ENIGMA Consortium. ENIGMA and global neuroscience: A decade of large-scale studies of the brain in health and disease across more than 40 countries. Transl Psychiatry 2020;10(01):100. Doi: 10.1038/ s41398-020-0705-1
  » https://doi.org/10.1038/s41398-020-0705-1
• ⁵⁷
  Sisodiya SM, Whelan CD, Hatton SN, et al; ENIGMA Consortium Epilepsy Working Group. The ENIGMA-Epilepsy working group: Mapping disease from large data sets. Hum Brain Mapp 2020;43 (01):113–128. Doi: 10.1002/hbm.25037
  » https://doi.org/10.1002/hbm.25037
• ⁵⁸
  Gleichgerrcht E, Munsell BC, Alhusaini S, et al; ENIGMA-Epilepsy Working Group. Artificial intelligence for classification of temporal lobe epilepsy with ROI-level MRI data: A worldwide ENIGMA-Epilepsy study. Neuroimage Clin 2021;31:102765. Doi: 10.1016/j.nicl.2021.102765
  » https://doi.org/10.1016/j.nicl.2021.102765
• ⁵⁹
  Louis S, Morita-Sherman M, Jones S, et al. Hippocampal Sclerosis Detection with NeuroQuant Compared with Neuroradiologists. AJNR Am J Neuroradiol 2020;41(04):591–597. Doi: 10.3174/ajnr. A6454
  » https://doi.org/10.3174/ajnr.A6454
• ⁶⁰
  Blumcke I, Spreafico R, Haaker G, et al; EEBB Consortium. Histopathological Findings in Brain Tissue Obtained during Epilepsy Surgery. N Engl J Med 2017;377(17):1648–1656. Doi: 10.1056/NEJMoa1703784
  » https://doi.org/10.1056/NEJMoa1703784
• ⁶¹
  Mellerio C, Labeyrie MA, Chassoux F, et al. 3T MRI improves the detection of transmantle sign in type 2 focal cortical dysplasia. Epilepsia 2014;55(01):117–122. Doi: 10.1111/epi.12464
  » https://doi.org/10.1111/epi.12464
• ⁶²
  Urbach H, Heers M, Altenmueller DM, et al. “Within a minute” detection of focal cortical dysplasia. Neuroradiology 2022;64 (04):715–726. Doi: 10.1007/s00234-021-02823-7
  » https://doi.org/10.1007/s00234-021-02823-7
• ⁶³
  Gill RS, Lee HM, Caldairou B, et al. Multicenter Validation of a Deep Learning Detection Algorithm for Focal Cortical Dysplasia. Neurology 2021;97(16):e1571–e1582. Doi: 10.1212/WNL.0000000000 012698
  » https://doi.org/10.1212/WNL.0000000000012698
• ⁶⁴
  Munsell BC, Wu G, Fridriksson J, et al. Relationship between neuronal network architecture and naming performance in temporal lobe epilepsy: A connectome based approach using machine learning. Brain Lang 2019;193:45–57. Doi: 10.1016/j. bandl.2017.08.006
  » https://doi.org/10.1016/j.bandl.2017.08.006
• ⁶⁵
  Cendes F, McDonald CR. Artificial Intelligence Applications in the Imaging of Epilepsy and Its Comorbidities: Present and Future. Epilepsy Curr 2022;22(02):91–96. Doi: 10.1177/15357597211068600
  » https://doi.org/10.1177/15357597211068600
• ⁶⁶
  Pardoe HR, Cole JH, Blackmon K, Thesen T, Kuzniecky RHuman Epilepsy Project Investigators. Structuralbrainchanges in medically refractory focal epilepsy resemble premature brain aging. Epilepsy Res 2017;133:28–32. Doi: 10.1016/j.eplepsyres.2017.03.007
  » https://doi.org/10.1016/j.eplepsyres.2017.03.007
• ⁶⁷
  Hwang G, Hermann B, Nair VA, et al. Brain aging in temporal lobe epilepsy: Chronological, structural, and functional. Neuroimage Clin 2020;25:102183. Doi: 10.1016/j.nicl.2020.102183
  » https://doi.org/10.1016/j.nicl.2020.102183
• ⁶⁸
  Morita-Sherman M, Li M, Joseph B, et al. Incorporation of quantitative MRI in a model to predict temporal lobe epilepsy surgery outcome. Brain Commun 2021;3(03):fcab164. Doi: 10.1093/braincomms/fcab164
  » https://doi.org/10.1093/braincomms/fcab164
• ⁶⁹
  Whiting AC, Morita-Sherman M, Li M, et al. Automated analysis of cortical volume loss predicts seizure outcomes after frontal lobectomy. Epilepsia 2021;62(05):1074–1084. Doi: 10.1111/epi.16877
  » https://doi.org/10.1111/epi.16877
• ⁷⁰
  Gleichgerrcht E, Munsell B, Bhatia S, et al. Deep learning applied to whole-brain connectome to determine seizure control after epilepsy surgery. Epilepsia 2018;59(09):1643–1654. Doi: 10.1111/epi.14528
  » https://doi.org/10.1111/epi.14528
• ⁷¹
  Thompson AJ, Banwell BL, Barkhof F, et al. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet Neurol 2018;17(02):162–173. Doi: 10.1016/S1474-4422(17)30470-2
  » https://doi.org/10.1016/S1474-4422(17)30470-2
• ⁷²
  Wingerchuk DM, Banwell B, Bennett JL, et al; International Panel for NMO Diagnosis. International consensus diagnostic criteria for neuromyelitis optica spectrum disorders. Neurology 2015;85 (02):177–189. Doi: 10.1212/WNL.0000000000001729
  » https://doi.org/10.1212/WNL.0000000000001729
• ⁷³
  Dos Passos GR, Oliveira LM, da Costa BK, et al. MOG-IgG- Associated Optic Neuritis, Encephalitis, and Myelitis: Lessons Learned From Neuromyelitis Optica Spectrum Disorder. Front Neurol 2018;9:217. Doi: 10.3389/fneur.2018.00217
  » https://doi.org/10.3389/fneur.2018.00217
• ⁷⁴
  Sati P, Oh J, Constable RT, et al; NAIMS Cooperative. The central vein sign and its clinical evaluation for the diagnosis of multiple sclerosis: a consensus statement from the North American Imaging in Multiple Sclerosis Cooperative. Nat Rev Neurol 2016;12(12):714–722. Doi: 10.1038/nrneurol.2016.166
  » https://doi.org/10.1038/nrneurol.2016.166
• ⁷⁵
  La Rosa F, Wynen M, Al-Louzi O, et al. Cortical lesions, central vein sign, and paramagnetic rim lesions in multiple sclerosis: Emerging machine learning techniques and future avenues. Neuroimage Clin 2022;36:103205. Doi: 10.1016/j.nicl.2022.103205
  » https://doi.org/10.1016/j.nicl.2022.103205
• ⁷⁶
  Kappos L, De Stefano N, Freedman MS, et al. Inclusion of brain volume loss in a revised measure of ’no evidence of disease activity’ (NEDA-4) in relapsing-remitting multiple sclerosis. Mult Scler 2016;22(10):1297–1305. Doi: 10.1177/13524585156 16701
  » https://doi.org/10.1177/1352458515616701
• ⁷⁷
  Cacciaguerra L, Storelli L, Rocca MA, Filippi M. Augmenting Neurological Disorder Prediction and Rehabilitation Using Artificial Intelligence. 6 - Current and future applications of artificial intelligence in multiple sclerosis. In: Pillai AS, Menon B, eds. Augmenting Neurological Disorder Prediction and Rehabilitation Using Artificial Intelligence. Academic Press; 2022:107–144
• ⁷⁸
  Cacciaguerra L, Meani A, Mesaros S, et al. Brain and cord imaging features in neuromyelitis optica spectrum disorders. Ann Neurol 2019;85(03):371–384. Doi: 10.1002/ana.25411
  » https://doi.org/10.1002/ana.25411
• ⁷⁹
  Rakić M, Vercruyssen S, Van Eyndhoven S, et al. icobrain ms 5.1: Combining unsupervised and supervised approaches for improving the detection of multiple sclerosis lesions. Neuroimage Clin 2021;31:102707. Doi: 10.1016/j.nicl.2021.102707
  » https://doi.org/10.1016/j.nicl.2021.102707
• ⁸⁰
  Aslam N, Khan IU, Bashamakh A, et al. Multiple Sclerosis Diagnosis Using Machine Learning and Deep Learning: Challenges and Opportunities. Sensors (Basel) 2022;22(20):7856 https://mdpi-res.com/d_attachment/sensors/sensors-22-07856/article_deploy/sensors-22-07856.pdf?version=1665912497 [Internet]
  » https://mdpi-res.com/d_attachment/sensors/sensors-22-07856/article_deploy/sensors-22-07856.pdf?version=1665912497
• ⁸¹
  Bonacchi R, Filippi M, Rocca MA. Role of artificial intelligence in MS clinical practice. Neuroimage Clin 2022;35:103065. Doi: 10.1016/j.nicl.2022.103065
  » https://doi.org/10.1016/j.nicl.2022.103065
• ⁸²
  Suh CH, Kim SJ, Jung SC, Choi CG, Kim HS. The "Central Vein Sign" on T2ω-weighted Images as a Diagnostic Tool in Multiple Sclerosis: A Systematic Review and Meta-analysis using Individual Patient Data. Sci Rep 2019;9(01):18188. Doi: 10.1038/s41598-019-54583-3
  » https://doi.org/10.1038/s41598-019-54583-3
• ⁸³
  Maggi P, Fartaria MJ, Jorge J, et al. CVSnet: A machine learning approach for automated central vein sign assessment in multiple sclerosis. NMR Biomed 2020;33(05):e4283. Doi: 10.1002/ nbm.4283
  » https://doi.org/10.1002/nbm.4283
• ⁸⁴
  Mendelsohn Z, Pemberton HG, Gray J, et al. Commercial volumetric MRI reporting tools in multiple sclerosis: a systematic review of the evidence. Neuroradiology 2023;65(01):5–24. Doi: 10.1007/s00234-022-03074-w
  » https://doi.org/10.1007/s00234-022-03074-w
• ⁸⁵
  Dillenseger A, Weidemann ML, Trentzsch K, et al. Digital Biomarkers in Multiple Sclerosis. Brain Sci 2021;11(11):1519. Doi: 10.3390/brainsci11111519
  » https://doi.org/10.3390/brainsci11111519
• ⁸⁶
  Sima DM, Esposito G, Van Hecke W, Ribbens A, Nagels G, Smeets D. Health Economic Impact of Software-Assisted Brain MRI on Therapeutic Decision-Making and Outcomes of Relapsing-Remitting Multiple Sclerosis Patients-A Microsimulation Study. Brain Sci 2021;11(12):1570. Doi: 10.3390/brainsci11121570
  » https://doi.org/10.3390/brainsci11121570
• ⁸⁷
  Nabizadeh F, Ramezannezhad E, Kargar A, Sharafi AM, Ghaderi A. Diagnostic performance of artificial intelligence in multiple sclerosis: a systematic review and meta-analysis. Neurol Sci 2023;44(02):499–517. Doi: 10.1007/s10072-022-06460-7
  » https://doi.org/10.1007/s10072-022-06460-7
• ⁸⁸
  Checkoway H, Lundin JI, Kelada SN. Neurodegenerative diseases. IARC Sci Publ 2011;(163):407–419
• ⁸⁹
  Young PNE, Estarellas M, Coomans E, et al. Imaging biomarkers in neurodegeneration: current and future practices. Alzheimers Res Ther 2020;12(01):49. Doi: 10.1186/s13195-020-00612-7
  » https://doi.org/10.1186/s13195-020-00612-7
• ⁹⁰
  Park M, Moon WJ, Structural MR. Structural MR Imaging in the Diagnosis of Alzheimer’s Disease and Other Neurodegenerative Dementia: Current Imaging Approach and Future Perspectives. Korean J Radiol 2016;17(06):827–845. Doi: 10.3348/kjr.2016. 17.6.827
  » https://doi.org/10.3348/kjr.2016.17.6.827
• ⁹¹
  Urs R, Potter E, Barker W, et al. Visual rating system for assessing magnetic resonance images: a tool in the diagnosis of mild cognitive impairment and Alzheimer disease. J Comput Assist Tomogr 2009;33(01):73–78. Doi: 10.1097/RCT.0b013e318163 73d8
  » https://doi.org/10.1097/RCT.0b013e31816373d8
• ⁹²
  Tăuţan AM, Ionescu B, Santarnecchi E. Artificial intelligence in neurodegenerative diseases: A review of available tools with a focus on machine learning techniques. Artif Intell Med 2021; 117:102081. Doi: 10.1016/j.artmed.2021.102081
  » https://doi.org/10.1016/j.artmed.2021.102081
• ⁹³
  Peralta C, Strafella AP, van Eimeren T, et al; International Parkinson Movement Disorders Society-Neuroimaging Study Group. Pragmatic Approach on Neuroimaging Techniques for the Differential Diagnosis of Parkinsonisms. Mov Disord Clin Pract (Hoboken) 2021;9(01):6–19. Doi: 10.1002/mdc3.13354
  » https://doi.org/10.1002/mdc3.13354
• ⁹⁴
  Schwarz ST, Afzal M, Morgan PS, Bajaj N, Gowland PA, Auer DP. The ’swallow tail’ appearance of the healthy nigrosome - a new accurate test of Parkinson’s disease: a case-control and retrospective cross-sectional MRI study at 3T. PLoS One 2014;9(04): e93814. Doi: 10.1371/journal.pone.0093814
  » https://doi.org/10.1371/journal.pone.0093814
• ⁹⁵
  Gaurav R, Yahia-Cherif L, Pyatigorskaya N, et al. Longitudinal Changes in Neuromelanin MRI Signal in Parkinson’s Disease: A Progression Marker. Mov Disord 2021;36(07):1592–1602. Doi: 10.1002/mds.28531
  » https://doi.org/10.1002/mds.28531
• ⁹⁶
  Gaurav R, Pyatigorskaya N, Biondetti E, et al. Deep Learning-Based Neuromelanin MRI Changes of Isolated REM Sleep Behavior Disorder. Mov Disord 2022;37(05):1064–1069. Doi: 10.1002/mds. 28933
  » https://doi.org/10.1002/mds.28933
• ⁹⁷
  Xu J, Zhang M. Use of Magnetic Resonance Imaging and Artificial Intelligence in Studies of Diagnosis of Parkinson’s Disease. ACS Chem Neurosci 2019;10(06):2658–2667. Doi: 10.1021/acschemneuro.9b00207
  » https://doi.org/10.1021/acschemneuro.9b00207
• ⁹⁸
  van der Burgh HK, Schmidt R, Westeneng HJ, de Reus MA, van den Berg LH, van den Heuvel MP. Deep learning predictions of survival based on MRI in amyotrophic lateral sclerosis. Neuroimage Clin 2016;13:361–369. Doi: 10.1016/j.nicl.2016.10.008
  » https://doi.org/10.1016/j.nicl.2016.10.008
• ⁹⁹
  Carrara G, Carapelli C, Venturi F, et al. A distinct MR imaging phenotype in amyotrophic lateral sclerosis: correlation between T1 magnetization transfer contrast hyperintensity along the corticospinal tract and diffusion tensor imaging analysis. AJNR Am J Neuroradiol 2012;33(04):733–739. Doi: 10.3174/ajnr.A2855
  » https://doi.org/10.3174/ajnr.A2855
• ¹⁰⁰
  Rajagopalan V, Chaitanya KG, Pioro EP. Quantitative Brain MRI Metrics Distinguish Four Different ALS Phenotypes: A Machine Learning Based Study. Diagnostics (Basel) 2023;13(09):1521. Doi: 10.3390/diagnostics13091521
  » https://doi.org/10.3390/diagnostics13091521

Publication Dates

• Publication in this collection
  28 June 2024
• Date of issue
  2024

History

• Received
  18 Oct 2023
• Accepted
  13 Dec 2023