Twenty-first century innovations in neuroradiology treatments

Abstract

Rapid technological progress in the 21^st century has transformed neuroradiology from primarily a diagnostic tool into a vital component of modern neuroscience. This narrative review summarizes innovations driving a paradigm shift in neurological and neurosurgical patient care. These advancements have introduced novel techniques and redefined models to improve short and long-term outcomes. Artificial intelligence (AI) has significant implications in stroke triage, brain tumor segmentation, and treatment planning. Wearable AI-enabled devices detect abnormal changes in movement and speech, prompting patients to seek care within the therapeutic window for reperfusion therapy. AI models can also identify tumors undetectable to the naked eye and provide precise lesion margins, reducing over- and underestimation during surgical planning. Image-guided minimally invasive procedures reduce reliance on open surgery while increasing precision, minimizing perioperative risk, shortening hospital stays, and improving outcomes. Refinements in stent and catheter technology enhance procedural safety and efficiency. Advanced imaging enables targeting of surgically inaccessible lesions and preservation of surrounding tissue. Innovations in neuroradiology address disparities in healthcare access in underserved regions and expand options for patient ineligible for open surgery. Continued advancements will optimize interventions, enhancing patient care. This review will help non-specialists in Neurology and Neuroradiology gain an overview of these innovations.

Key Words: Neuroradiology; Imaging; Minimally invasive; Artificial intelligence; Intracranial; Endovascular

Core Tip: There have been considerable advancements in neuroradiological treatments over recent decades, particularly in the 21^st century. This review aims to comprehensively examine the evolution of neuroradiology across multiple domains, including cerebrovascular accidents, intracranial endovascular techniques, brain tumors, spine interventions, and image-guided therapeutic strategies for neurological disorders. Attention is particularly given to evolving innovations, such as artificial intelligence-driven diagnostics, minimally invasive endovascular and spine interventions, and gene and cellular therapies. Moreover, this review evaluates the impact of these developments on patient outcomes, including improved prognosis, reduced hospital length of stay, and enhanced quality of life.

INTRODUCTION

The 20^th century marked a period of exploration in neuroradiology, defined by the development of foundational imaging techniques and interventional tools. These early innovations revealed possibilities that fueled 21^st-century advances focused on redefining existing methods, integrating emerging technologies, optimizing techniques, and improving precision and patient safety. Minimally invasive therapy, the expanding use of artificial intelligence (AI), and advances in imaging have since revolutionized the management of neurological diseases. Recent developments in computed tomography (CT) and magnetic resonance imaging (MRI) technology now allow clinicians to visualize neurovascular structures with much greater clarity than ever before[1]. Minimally invasive treatments such as endovascular embolization have been shown to improve the safety and effectiveness of arteriovenous malformation (AVM) treatment[2]. Concurrently, AI has helped generate solutions to complex neurological problems, leading to the development of algorithms that can be applied in clinical practice[3].

Despite these major advancements, there is limited consolidated literature that evaluates how these innovations translate into improved clinical outcomes, reduced procedural risks, and enhanced quality of life for patients. This may partly explain why deep learning, although very successful in controlled research environments, has appeared in only a small number of clinical trials[4]. While the information exists, organizing it in a centralized, accessible format may facilitate clinicians’ engagement, increase awareness of emerging tools, and encourage exploration of new approaches in clinical practice.

A broad overview of these innovations is therefore essential to identify the current strengths and limitations in neuroradiology. Such an assessment can highlight areas of the field that require further investigation and help guide future research priorities. Given the rapid advancements in technology, continuous reassessment is necessary to ensure that clinical practice employs evidence-based tools and treatments. In this review, we discuss 21^st-century advances in minimally invasive therapies, AI, and imaging in neuroradiology to illustrate progress, identify future priorities, and provide neurologists, internists, intensivists, hospitalists, allied practitioners, nurses, and clinicians with practical insights to improve patient care.

LITERATURE SEARCH

A focused literature search was conducted from November 23, 2025, through December 12, 2025, using PubMed, ScienceDirect, JSTOR, and Google Scholar. Search strategies combined key terms related to AI, minimally invasive, endovascular, and neuroimaging with condition and procedure-specific keywords such as mechanical thrombectomy, flow diverters, laser interstitial thermal therapy (LITT), and focused ultrasound. Searches were performed using combinations of these terms, and additional relevant articles were identified through manual review of reference lists from selected studies.

In this narrative review, articles were selected based on their clinical relevance to minimally invasive techniques, AI, and advanced imaging in neuroradiology. Only peer-reviewed publications from 1990 onward were included to capture developments reflective of modern neuroradiological practice. Both original research articles and review articles were considered. Articles published before 1990, non-peer-reviewed sources, and studies that were not relevant to the scope of this review were excluded. The study selection was guided by thematic relevance, with emphasis on literature representing key innovations in the field. Thematic relevance was determined based on whether an article substantially contributed to the discussion of the emerging technologies, diagnostic advancements, or therapeutic innovations in neuroradiology. Using these criteria, a total of 66 peer-reviewed articles were identified and included.

The literature was synthesized thematically and organized into major categories and subcategories. Subcategories included mechanical thrombectomy for acute ischemic stroke, AI-driven stroke triage and imaging analysis, endovascular aneurysm treatment using flow diverters, minimally invasive tumor ablation with LITT, focused ultrasound for essential tremor and Parkinson’s disease, and endovascular treatment of AVMs. This narrative review was conducted in accordance with established principles for narrative synthesis in medical literature. An earlier version of this work was made available as a preprint without peer review and has since been revised and updated for submission to this journal[5].

ADVANCES IN MANAGEMENT OF CEREBROVASCULAR ACCIDENTS

Stroke triage and imaging analysis

AI has transformed modern medicine by introducing novel approaches to disease diagnosis, reducing diagnostic errors, accelerating drug discovery, enhancing communication between clinicians and patients, and enabling rapid analysis of medical imaging. Among its most impactful applications has been its integration into neuroradiology. In the 21^st century, AI has significantly improved stroke triage and imaging interpretation. Both traditional machine learning techniques and deep learning models have been applied across multiple domains of stroke management, including stroke detection, identification of large vessel occlusions, assessment of perfusion deficits, estimation of time since symptom onset, and prediction of functional recovery[6]. Collectively, these advances demonstrate the role of AI in enabling more timely and accurate patient evaluation, with direct implications for clinical decision-making and patient outcomes.

Timely recognition remains one of the most critical determinants of outcome in acute ischemic stroke. Earlier identification of neurological symptoms allows patients to seek care sooner, undergo diagnostic imaging, and receive appropriate intervention. However, a retrospective study reported that 74.6% of patients arrived more than 4.5 hours after symptom onset, with only 22% arriving within three hours[7]. Given that intravenous tissue plasminogen activator is limited to a 4.5-hour therapeutic window, most patients present too late to qualify for reperfusion therapy. AI-enabled technologies including smartphones, wearable sensors, smartwatches, and tablets, have therefore emerged as tools to reduce prehospital delays by monitoring changes in motor function and speech. A prospective study demonstrated that wrist-worn accelerometers integrated with AI algorithms could detect unilateral upper extremity weakness when worn continuously, achieving high diagnostic accuracy with an area under the curve ranging from 0.893 to 0.947 for detection windows as short as 15 minutes after symptom onset[8]. Similarly, a smartphone-based application using machine learning algorithms analyzed 18311 images and detected facial asymmetry with a sensitivity of 99.42%, specificity of 93.67%, and overall accuracy of 97.11%[9]. Together, these technologies highlight a consistent trend toward enhancing prehospital stroke triage and facilitating the timely routing of patients to appropriate stroke centers.

AI platforms have also been implemented for the rapid detection of large vessel occlusions. Automated software such as RAPID CT angiography correctly identified 93% of patients with occlusions, achieving a sensitivity of 0.94 and a negative predictive value of 0.98[10]. Similarly, the Food and Drug Administration (FDA)-cleared Viz.ai convolutional neural network demonstrated high diagnostic performance with rapid processing times of approximately five minutes[11]. When combined with CT perfusion imaging, AI-based tools can estimate ischemic core and penumbral tissue volumes, allowing neurologists to expedite interventions such as mechanical thrombectomy[6]. In several studies, AI-based imaging tools consistently demonstrate high diagnostic accuracy and rapid processing times, supporting their potential role in accelerating treatment decisions. Given that large vessel occlusions account for a substantial proportion of stroke-related disability and mortality, reductions in prehospital delays and treatment times may significantly improve outcomes and reduce the global burden of disease.

Despite these advances made with AI, there are several limitations in its application to clinical practice. Many models are developed in specific environments and may only be tested within those settings, which raises concerns about their generalizability across different patient populations and clinical settings. Using these models without evaluating them in broader populations may lead to delays in care or less effective clinical decisions. Dataset bias may also affect performance, where models work well in one group but not in others, potentially contributing to disparities in care. As many models have not undergone external validation, there is limited confidence in their reliability in real-world clinical settings. In addition, because AI models depend on human-labeled data, data annotation bias can affect performance, meaning that errors or differences in how the data are labeled can be learned by the model. As a result, models trained on different datasets may perform differently, which can further affect reliability. Given that many AI systems function as “black box” models, where the reasoning behind their decisions is not easily understood, this may limit clinical trust and prevent adoption into real-world practice, especially in a field where physicians are responsible for patient outcomes.

Mechanical thrombectomy for acute ischemic stroke

Mechanical thrombectomy is a minimally invasive procedure that uses catheter-based techniques, including stent retrievers or aspiration catheters, to remove intravascular clots under image guidance. It is primarily indicated for acute ischemic stroke, a leading cause of death and disability worldwide[12]. For decades, intravenous thrombolysis was the most commonly used treatment for acute ischemic stroke; however, limitations such as a narrow therapeutic window for tissue plasminogen activator administration (3-4.5 hours), low recanalization rates for large vessel occlusions, and numerous contraindications, including recent surgery and active bleeding, restricted its overall effectiveness[13]. Advances in mechanical thrombectomy during the 21^st century have shifted standards of care, establishing it as first-line therapy for eligible patients.

These improvements are consistently observed across studies, particularly in patients with large vessel occlusions. Clinical studies demonstrate that early thrombectomy using devices such as the Solitaire FR stent retriever results in higher rates of reperfusion, faster neurologic improvement, and superior functional outcomes compared with intravenous alteplase alone[14]. Importantly, mechanical thrombectomy has proven effective for large clots and may be performed up to 24 hours after symptom onset in select patients. In a landmark randomized controlled trial, patients who received endovascular therapy achieved significantly greater reperfusion than those treated with alteplase alone, with neurologic improvement observed within three days and improved functional outcomes at 90 days[15].

Beyond clinical recovery, image-guided removal of large vessel occlusions has had broad implications for survival, disability, and healthcare utilization. Campbell et al[14] demonstrated that patients treated with mechanical thrombectomy spent significantly more days at home within the first 90 days after discharge compared with those receiving alteplase alone (median 73 days vs 15 days; P = 0.001), fewer days in acute stroke units (median 5 days vs 8 days; P = 0.04), and fewer days in rehabilitation facilities (mean 14 days vs 33 days). Additionally, mechanical thrombectomy reduced mortality risk and yielded a median gain of 9.3 disability-adjusted life years, compared with 4.9 in the control group. However, the follow-up period in this study was limited to 90 days, restricting assessment of long-term functional outcomes. Furthermore, the findings may not be generalizable to patients with pre-existing disabilities, as such individuals were excluded from the study population. The study was also conducted in a single-center setting, which may further limit the external validity of the findings. Together, these limitations may potentially reduce applicability to real-world populations with more diverse clinical and demographic characteristics. Additionally, one related trial evaluating similar interventions was terminated early, which may affect the robustness and interpretability of its results[15]. Higher upfront procedural cost also remains a consideration and in future studies, it should include a cost-benefit analysis to better evaluate their economic feasibility across different healthcare systems. Nonetheless, mechanical thrombectomy’s substantial gains in quality-adjusted life years and reductions in inpatient care make it a cost-effective intervention (Figure 1).

Figure 1  Mechanical thrombectomy for acute ischemic stroke.

ADVANCES IN THE INTRACRANIAL-ENDOVASCULAR TECHNIQUES

Endovascular stenting for intracranial atherosclerotic disease

Early-generation stents were moderately effective but carried substantial risks due to large cell openings, excessive rigidity, limited conformability, poor scaffolding coverage, incompatibility with embolic protection devices, and increased endothelial injury. For example, bare-metal stents are prone to neointimal hyperplasia, which can lead to restenosis and increase the risk of ischemic stroke[16]. Drug-eluting stents (DES) were introduced to address this limitation. A systematic review comparing DES with bare-metal stents demonstrated significantly lower rates of in-stent restenosis and stroke recurrence within one year among patients treated with DES[17]. In the 21^st century, continued refinements in stent design have made endovascular treatment safer and more effective. However, Khan et al[18] found that intracranial stenting was more costly than medical therapy for intracranial stenosis, with similar quality-adjusted life-year (QALY) between groups. The cost per QALY gained was US$ 20542 for stenting vs US$ 4265 for medical management, indicating that cost-effectiveness is more likely in patients with symptomatic stenosis ≥ 70%.

Modern stents are engineered to overcome the limitations of earlier devices. Dual-layer micromesh stents feature micron-sized pores that prevent plaque protrusion and function as a protective net, trapping debris that could otherwise embolize during stent deployment, post-dilatation, and the early post-stenting period[19]. This design is associated with reduced rates of major adverse events, including stroke and death. A retrospective study found that patients undergoing carotid artery stenting with single-layer stents experienced significantly more periprocedural neurological complications, particularly transient ischemic attacks (TIA), than those treated with dual-layer micromesh stents[20], although these findings should be interpreted with caution given the retrospective design and potential limitations in sample size and follow-up duration. Balloon-expandable stents provide precise deployment and high radial force, enabling immediate, stable expansion and facilitating treatment of rigid, calcified plaques. In patients with intracranial arterial stenosis, Qureshi et al[21] reported that balloon-expandable stents were associated with significantly lower residual postprocedural stenosis and reduced rates of postprocedural stroke and death compared with self-expanding stents. In contrast, Zaidat et al[22] demonstrated higher 12-month rates of stroke or TIA in the same territory and increased 30-day risk of any stroke or TIA with balloon-expandable intracranial stents compared with medical therapy in symptomatic intracranial arterial stenosis. These conflicting findings highlight the uncertainty regarding the long-term superiority of balloon-expandable stenting over optimized medical management and suggest the need for cautious patient evaluation and further long-term comparative investigation.

In addition to device refinement, improved patient selection criteria have strengthened the role of endovascular therapy. Historically, stents were used too broadly, leading to unfavorable outcomes. Clinical trial data have clarified which patients benefit most from stenting and which are at increased risk. For example, middle cerebral artery stenting in elderly patients has been associated with a high risk of ischemic stroke[23], whereas younger patients generally have more favorable vascular anatomy and fewer complications. Endovascular stenting is not recommended for patients with stroke attributable to 50%-69% intracranial arterial stenosis, as medical therapy alone is associated with a low recurrence rate[24]. Appropriate candidates for stenting include patients with high-grade stenosis, severe cardiopulmonary disease, renal failure, or prior neck radiation. Careful patient selection allows stenting to be reserved for those most likely to benefit while reducing unnecessary procedures, complications, and healthcare costs. A summary of these modalities is included in Table 1.

Table 1 Comparison of endovascular stents.

Feature	Early-generation/bare-metal stents	Drug-eluting stents	Modern stents (micromesh/balloon-expandable)
Design	Large cell openings, rigid, limited conformability	Drug-eluting stents	Micron-sized pores (micromesh); high radial force (balloon-expandable)
Mechanism	Mechanical scaffolding	Reduces neointimal hyperplasia	Prevents plaque protrusion; traps debris; improves vessel expansion
Key outcomes	Associated with restenosis and increased stroke risk	Lower rates of in-stent restenosis and stroke recurrence	Lower residual stenosis and fewer complications (vs single-layer stents)
Complications	Endothelial injury; restenosis		Stroke/TIA risk varies depending on the comparison (vs self-expanding stents or medical therapy)
Evidence comparison		Better outcomes than bare-metal stents	Qureshi: Improved outcomes vs self-expanding stents; Zaidat: Worse outcomes vs medical therapy
Cost consideration			Higher cost than medical therapy; more favorable in ≥ 70% stenosis

TIA: Transient ischemic attacks.

Endovascular aneurysm treatment (flow diverters)

Flow-diverting stents represent a major advancement in the endovascular management of intracranial aneurysms, redirecting blood flow away from the aneurysm sac to promote gradual thrombosis and healing without the need for open surgery. Early attempts to treat intracranial aneurysms, while often beneficial, were associated with substantial limitations and procedural risks. Surgical ligation of the parent artery effectively occluded aneurysms but disrupted distal blood flow, increasing the risk of ischemic stroke. Subsequent techniques, including surgical clipping and endovascular coil embolization, addressed this limitation by isolating the aneurysm sac while preserving the parent vessel. Despite their widespread use, these approaches remain suboptimal for large, wide-necked, or fusiform aneurysms[25].

The development of flow-diverting stents in the 21^st century marked a paradigm shift in aneurysm management. Unlike traditional intracranial stents, which are primarily designed to support coils, flow diverters alter intra-aneurysmal hemodynamics to promote progressive occlusion and parent vessel reconstruction[26]. These devices redirect blood flow away from the aneurysm, inducing thrombosis within the sac while maintaining patency of the parent artery. Key design features include low porosity, increased metal surface area coverage, and high radial force, which together stabilize the construct, reduce aneurysmal inflow, and promote durable vessel remodeling.

A major advantage of flow diversion is its minimally invasive nature, which eliminates the need for craniotomy, reduces perioperative risk, and shortens recovery time. Compared with earlier techniques, flow diversion has demonstrated lower recurrence rates and higher rates of complete occlusion in several studies. One study reported complete occlusion in 55.1% of aneurysms treated with stent-assisted coiling, with a recurrence rate of 28%, whereas 86.7% of aneurysms treated with flow diverters achieved complete occlusion with only a 2.2% recurrence rate[27]. These outcomes reflect the ability of flow diverters to reconstruct the parent artery and achieve durable aneurysm exclusion. However, the study reporting favorable outcomes with flow-diverter stenting possesses methodological limitations, as it was a non-randomized, single-center investigation with a relatively small patient cohort, which may introduce selection bias and limit generalizability to broader patient populations.

Flow diversion was specifically developed to address complex aneurysms poorly suited to earlier approaches. In wide-necked aneurysms, coil embolization carries a risk of coil prolapse into the parent artery, potentially leading to thromboembolism, whereas surgical clipping can be challenging due to the risk of clip slippage or incomplete occlusion. In a clinical trial of 108 patients with unruptured large and giant wide-necked aneurysms, complete occlusion was achieved in 73.6% of cases at six months and 86.8% at 12 months[28]. Although flow-diverting stents have demonstrated promising results in the management of intracranial aneurysms, several potential complications remain a concern. Reported adverse events include ipsilateral ischemic strokes, intraparenchymal hemorrhage, subarachnoid hemorrhage, increased intra-aneurysmal pressure, and thrombus-associated autolysis of the aneurysm[26]. Despite these limitations, including the risk of in-stent thrombosis and the need for prolonged dual antiplatelet therapy, continued refinement has expanded its safety profile and therapeutic applications. As a result, flow-diverting stents have become an integral and evolving tool in contemporary aneurysm management.

Endovascular treatment of AVMs

Intracranial AVMs are abnormal vascular connections that can result in intracerebral hemorrhage, seizures, and vascular steal phenomena, all of which may be life-threatening. Open surgical resection was the earliest treatment modality developed for AVMs and remains the gold standard for select patients, owing to its high obliteration rates[29]. Microsurgical series have reported complete AVM obliteration rates exceeding 98%, with favorable functional outcomes achieved in the majority of patients[30]. However, surgical resection is limited in deep-seated or complex AVMs, where operative access is challenging and hemorrhagic risk is substantial.

Endovascular therapy emerged in the 1970s as a minimally invasive alternative involving catheter-based delivery of embolic agents to reduce or eliminate flow within the AVM nidus. Although embolization can serve as definitive therapy in select cases, it has evolved primarily as an adjunct to microsurgery, improving operative safety and efficacy[2]. Advances in embolic materials during the 21^st century, most notably the introduction of agents such as Onyx and PHIL, have further improved treatment outcomes.

Early embolic agents, including n-butyl cyanoacrylate (n-BCA), played a critical role in facilitating surgical resection by reducing nidus size and delineating arterial feeders. However, n-BCA polymerizes rapidly upon contact with blood, making injection control difficult and increasing the risk of catheter entrapment or vessel occlusion. In contrast, non-adhesive agents such as Onyx and PHIL solidify more slowly, allowing prolonged and controlled injection with deeper nidus penetration. A retrospective study demonstrated significantly higher occlusion rates and reduced need for subsequent surgery with Onyx compared with n-BCA[31]. These findings suggest that, compared with early embolic agents, modern embolic agents provide improved injection control and higher rates of effective nidus occlusion, although this conclusion may be limited by the study design and may not fully reflect long-term outcomes.

Advances in microcatheter technology have further enhanced procedural safety. Early catheters were prone to entrapment due to adhesive embolic agents and limited flexibility, increasing the risk of arterial perforation[32]. The introduction of detachable tip microcatheters addressed this limitation by allowing the distal segment to detach safely when necessary. The Apollo microcatheter, introduced in 2014, was the first commercially available detachable-tip system. In a prospective post-market study, catheter tip detachment occurred intentionally in over half of procedures, with minimal unintended detachment and no associated complications[33]. However, limitations such as premature tip detachment remain a concern. Potential complications such as vessel occlusion, thromboembolism, vessel injury, and the need for retrieval procedures are likely. These complications may increase patient morbidity, prolong recovery time, and increase healthcare costs for both the patient and hospital. Nonetheless, these innovations have improved control, safety, and efficacy in modern endovascular AVM treatment.

Neurovascular robotics and remote intervention

Robotic systems have been widely adopted across multiple medical specialties; however, their application in neuroendovascular interventions has only been reported since the early 2000s. Prior to that time, physicians relied solely on traditional methods and manual dexterity, which were associated with increased radiation exposure, longer procedure times, and possible hand tremors that could reduce procedural precision. Neurovascular robotics aim to address these limitations. In 2007, a feasibility study demonstrated that magnetic navigation systems combined with magnetic micro guidewires enabled safe and accurate microcatheter placement in patients with neurovascular disease, with no reported complications or deaths[34]. Modern endovascular robotic platforms consist of two primary components: A patient-side robotic unit that manipulates catheters and guidewires, and a remote-control station where operators direct device movement using joysticks, sensors, and computer interfaces[35]. These systems enable precise catheter navigation and open the door to remote neurointerventional procedures, potentially reducing disparities in access to specialized stroke care.

Given the time-sensitive nature and consequences of conditions such as acute ischemic stroke and large aneurysms, neurovascular robotics holds clinical promise. In a study involving four remote clinicians and one on-site clinician connected to a robotic system, operators successfully navigated catheters and guidewires from the femoral artery to the middle cerebral artery within 15 minutes without manual assistance[36]. In 2019, a patient with a wide-necked (> 10 mm) saccular sidewall aneurysm of the distal basilar artery underwent successful robotic-assisted aneurysm coiling and stent placement using the CorPath GRX Robotic System, achieving complete obliteration. The procedure was completed in 2 hours and 9 minutes. Across these studies, neurovascular robotics demonstrate feasibility and safety while also highlighting the potential for remote systems to enable timely intervention in hospitals in rural or underserved regions without on-site neurointerventionalists.

Despite these advances, current robotic systems face technical limitations. The CorPath GRX system, for example while FDA-approved for coronary and peripheral interventions, lacks support for the triaxial catheter approach required for most neurovascular procedures[35]. In addition, the loss of tactile feedback raises concerns about vessel perforation since physicians cannot feel resistance during the procedure. Furthermore, the high equipment cost may limit its widespread adoption. Nonetheless, as robotic platforms continue to evolve and adapt to neurovascular demands, these technologies may overcome geographic barriers, expedite stroke treatment, and improve patient outcomes.

ADVANCES IN MANAGEMENT OF BRAIN TUMORS

Minimally invasive tumor ablation LITT

Minimally invasive tumor ablation is an image-guided approach that uses thermal energy, cryoablation, or targeted energy delivery to destroy tumor tissue. These techniques are particularly well-suited for small lesions that are difficult to access surgically and for patients who are poor candidates for open procedures. Among available modalities, LITT has emerged as a promising option. Although LITT was first introduced in 1983, early adoption was limited by the inability to monitor tissue temperature during laser application, which restricted precise control of ablation extent[36,37]. Advances in intraoperative MRI during the 21^st century have renewed interest in LITT by enabling real-time thermometry using T1-weighted two-dimensional images acquired throughout the procedure. This capability allows clinicians to monitor thermal spread and minimize injury to adjacent structures precisely. With improvements in safety and precision, LITT has become an established treatment option for epilepsy, radiation necrosis, and metastatic brain tumors while minimizing collateral damage.

Technological refinements have ensured that thermal injury is largely confined to target tissue while sparing surrounding brain structures[38]. This precision has been especially beneficial in managing lesions considered high risk or surgically inaccessible using conventional approaches. Currently, two major commercially available systems exist: NeuroBlate and Visualase. NeuroBlate offers directional laser capabilities advantageous for irregularly shaped lesions, whereas Visualase is better suited for symmetric targets. In a study using the NeuroBlate system, median survival in patients with recurrent glioblastoma increased from 90-150 days to 361 days following LITT[39]. Additional evidence supports the efficacy of LITT in metastatic brain disease. In a study of patients with recurrent metastatic lesions previously treated with radiation therapy, LITT was associated with minimal discomfort, early discharge, and radiographic evidence of effective ablation[40]. Subsequent studies reported that 60%-75% of treated metastatic lesions did not recur within six months. The consistent trend in these studies suggests that LITT provides effective local tumor control with favorable recovery profiles.

LITT has also been incorporated into the management of medically refractory epilepsy. In a cohort of 77 patients with drug-resistant mesial temporal lobe epilepsy, 58% achieved Engel class I outcomes at two years, and 57% achieved International League Against Epilepsy class 1 or 2 outcomes following LITT[41]. These outcomes are comparable to those of anterior temporal lobectomy while offering a substantially less invasive alternative. However, limitations remain, including postprocedural cerebral edema, reduced effectiveness for large lesions, and contraindications to patients who cannot undergo MRI. Specifically, postprocedural edema is a concern because of its potential consequences, including increased intracranial pressure, neurological deficits such as visual changes, sensory deficits, and cognitive impairment, seizures, and brain herniation. These complications may require additional interventions such as corticosteroids, repeat imaging, or decompressive therapy, which can prolong hospitalization and recovery. As imaging technologies and image-guided therapeutic systems continue to advance and the clinical applications of LITT expand further, these limitations should be taken into consideration to support safer clinical integration and improve long-term therapeutic outcomes.

Brain tumor segmentation and treatment planning

Cancer remains the second leading cause of death in the United States, and brain tumors pose unique diagnostic and therapeutic challenges[42]. The integration of AI into oncology has begun to address limitations in tumor detection, characterization, and treatment planning that have historically relied on manual interpretation[43]. In neuro-oncology, AI has enhanced the precision and efficiency of brain tumor management; however, several limitations remain, including ethical, legal, and social considerations that reflect broader limitations in the clinical application of AI, including concerns related to data quality and model interpretability. Ethical concerns, such as patient privacy and the need for transparency and accountability in AI systems, are essential to sustain patient trust. Legal challenges, including liability for AI-related errors and medical malpractice, must also be addressed. Social implications, such as effects on the patient-physician relationship, patient autonomy, and healthcare disparities, also warrant careful consideration[44]. For example, reliance on AI assisted systems can influence communication and shared clinical decision-making. As these technologies continue to evolve, clearer standards for clinical accountability will be necessary to support safe and ethical implementation and maintain patient trust.

AI-based models can detect small lesions that may be overlooked on conventional imaging. In a study evaluating a three-dimensional U-Net convolutional neural network for detecting brain lesions on ¹⁸F- fluoro-ethyl-L-tyrosine positron emission tomography (PET) imaging, the model achieved an accuracy of 0.9868 during training and 0.9856 during validation, with 100% sensitivity and specificity and no false positives[45]. Early detection enables earlier intervention, potentially preventing tumor progression and improving prognosis.

Accurate tumor segmentation is essential for diagnosis, longitudinal monitoring, and treatment planning. Manual segmentation is time-consuming and subject to interobserver variability[46]. AI-based segmentation improves efficiency and reproducibility. In a study using two-dimensional U-Net models trained on expert-labeled data, mean Sørensen-Dice scores of 0.80, 0.84, and 0.91 were achieved for enhancing tumor, tumor core, and whole tumor segmentation, respectively[47]. These improvements reduce the risk of overtreatment and enable precise surgical and radiation planning. By providing objective quantification of tumor boundaries and treatment response, AI supports more precise treatment planning, enabling earlier intervention and improved clinical outcomes.

Beyond tumor detection and segmentation, machine learning models have also demonstrated utility in glioma imaging biomarkers for diagnosis, prognosis, and treatment response monitoring. Recent studies have also explored additional applications of machine learning and neural networks in neuro-oncology, including radiomic analysis and personalized treatment strategies. However, many of these studies remain retrospective in design, emphasizing the need for larger multicenter validation studies before broader clinical implementation. Nonetheless, these expanding applications further highlight the growing role of AI-assisted systems in neurodiagnostic and neurointerventional practice[44,48].

Advanced perfusion and functional imaging integration

Historically, neurosurgical planning relied heavily on anatomical landmarks and intraoperative cortical stimulation, limiting preoperative precision and increasing the risk of neurological deficits. Critical white matter pathways, including the corticospinal tract and arcuate fasciculus, were invisible on conventional MRI, making them vulnerable to inadvertent injury during tumor resection. In the 21^st century, advanced imaging technologies have transformed neurosurgical practice by enabling precise localization of pathology and preservation of the eloquent cortex[49].

Advanced perfusion and functional imaging modalities, including CT/MRI perfusion, diffusion tensor imaging (DTI), and functional MRI (fMRI), now work synergistically to generate detailed functional and vascular maps. Dynamic susceptibility contrast MRI measures cerebral blood flow, cerebral blood volume, and mean transit time through contrast-induced signal changes on T2-weighted images[50]. A study by Hakyemez et al[51] demonstrated that relative cerebral blood volume ratios were significantly higher in high-grade gliomas than in low-grade tumors, enabling reliable tumor grading.

DTI further enhances surgical planning by visualizing white-matter tract orientation and integrity, enabling surgeons to assess tumor displacement or infiltration[52]. When combined with fMRI, which identifies eloquent cortical regions using blood oxygen level–dependent signals, these modalities provide comprehensive structural and functional mapping. Case studies have demonstrated that combined DTI and fMRI enable identification of functional cortex and displaced tracts, contributing to favorable postoperative neurological outcomes[53]. Together, these tools support maximal safe resection while preserving neurological function.

ADVANCES IN SPINE INTERVENTIONS

Spine cement augmentation (vertebroplasty and kyphoplasty)

Vertebral compression fractures are disruptions of the vertebral body that commonly result in severe back pain involving the thoracic and lumbar spine. Although fractures may arise from trauma, infection, or malignancy, osteoporosis remains the most common underlying cause. Osteoporosis is characterized by reduced bone density and structural deterioration, increasing the risk of fractures. It is estimated that approximately 1.4 million osteoporotic vertebral compression fractures occur annually, and nearly 40% of women experience one during their lifetime[54]. Historically, conservative management such as bed rest, bracing, analgesics, physical therapy, and nutritional supplementation served as the primary treatment strategy. However, these approaches do not stabilize fractures or restore vertebral body integrity. In the 21^st century, image-guided vertebral augmentation procedures, specifically vertebroplasty and kyphoplasty, have emerged as effective minimally invasive interventions that rapidly improve pain and mobility, particularly in osteoporotic patients[55].

Vertebroplasty restores spinal mechanics by stabilizing a fractured vertebral body through percutaneous injection of polymethylmethacrylate cement under fluoroscopic or CT guidance. Cement hardening reinforces the vertebra, reduces micromotion, and alleviates pain[56]. In a retrospective study of 38 patients with 70 symptomatic osteoporotic fractures refractory to medical therapy, 68% experienced complete pain relief within 48 hours, while 32% reported moderate improvement[57]. At a mean follow-up of 18 months, 94% of patients reported sustained pain relief, with a low complication rate of 6.4%.

Kyphoplasty similarly employs image-guided cement injection but incorporates a balloon-tamponade device to create a cavity within the vertebral body before cement placement. This technique restores vertebral height and reduces kyphotic deformity. Kyphoplasty has been shown to restore up to 97% of vertebral height compared with approximately 30% achieved with vertebroplasty alone[58]. The balloon also reduces cement extravasation by sealing potential leakage pathways. A systematic review demonstrated significantly greater pain reduction at one month and sustained improvement at three months compared with conservative management[58]. However, these findings are not entirely consistent across studies. Lee et al[59] found that 8 of 19 reported no significant differences in clinical outcomes between kyphoplasty and conservative management at 3, 6, or 12 months, except at 1 month; therefore, kyphoplasty may be considered within the first month and limited to selected patients, aged > 78.5 years, with a T-score < -2.95, BMI > 25.5 kg/m², and vertebral collapse > 28.5%. Similarly, Yi et al[48] reported that new vertebral fractures occurred in both treatment groups during long-term follow-up (36-80 months), but fracture onset occurred earlier following cement augmentation than with conservative therapy. Although both procedures carry risks, including cement leakage, adjacent-level fractures, and postprocedural pain, their favorable safety profiles and demonstrated functional benefits support their role in the management of vertebral compression fractures.

Minimally invasive spine interventions

Minimally invasive spine interventions treat spinal disorders such as disc herniation and spinal stenosis through small incisions with minimal disruption of surrounding tissue. Compared with traditional open surgery, these procedures reduce tissue trauma, blood loss, hospital stay, and recovery time while maintaining safety and efficacy[60]. Advances in imaging and instrumentation during the 21^st century have transformed procedures such as epidural steroid injections, percutaneous discectomy, and spinal ablation into precise therapeutic options.

Epidural steroid injections deliver corticosteroids into the epidural space to reduce inflammation and relieve radicular pain. Advances in C-arm fluoroscopy and transforaminal techniques have improved targeting accuracy. In patients with cervical radicular pain refractory to conservative therapy, transforaminal epidural steroid injections performed using advanced fluoroscopic systems resulted in significant improvements in PROMIS Pain Interference scores at 3, 6, and 12 months[61].

Percutaneous discectomy removes herniated disc material using image-guided instruments without general anesthesia, reducing procedural risk. Advances in endoscopic visualization have improved precision and outcomes. Compared with open discectomy, percutaneous endoscopic lumbar discectomy has been associated with reduced blood loss, shorter hospitalization, smaller incisions, and lower postoperative inflammatory markers[62]. Adjunctive spinal ablation further enhances outcomes. Patients undergoing combined percutaneous discectomy and sinuvertebral nerve ablation experienced significantly lower pain and disability scores compared with discectomy alone[63]. These findings suggest that minimally invasive spine interventions provide effective treatment options for patients with lumbar disc herniation and chronic low back pain with reduced procedural morbidity compared with traditional open surgery. A summary of these interventions is included in Tables 2 and 3.

Table 2 Comparison of endovascular techniques.

Feature	Microsurgical resection	Early embolic agents (n-BCA)	Modern embolic agents (Onyx, PHIL)
Role	Gold standard for select patients	Adjunct to surgery; can be definitive in select cases	Adjunct or definitive therapy in select cases
Invasiveness	Open surgery	Minimally invasive	Minimally invasive
Mechanism	Direct removal of AVM nidus	Rapid polymerization upon contact with blood	Slow solidification allowing controlled injection and deeper penetration
Effectiveness	High obliteration rates (> 98%)	Used to reduce nidus size and delineate feeders	Higher occlusion rates and reduced need for surgery
Injection control		Difficult due to rapid polymerization	Improved control with prolonged injection
Safety concerns	Limited in deep/complex AVMs; hemorrhagic risk	Catheter entrapment, vessel occlusion	Improved control; safer delivery profile
Technological features		Adhesive embolic agent	Non-adhesive agents with deeper nidus penetration
Microcatheter considerations		Risk of entrapment with early systems	Detachable-tip microcatheters improve safety
Limitations	Limited in deep-seated or complex AVMs	Poor injection control and higher complication risk	Risk of premature tip detachment

nBCA: N-butyl cyanoacrylate; PHIL: Precipitating hydrophobic injectable liquid; AVM: Arteriovenous malformation.

Table 3 Minimally invasive spine interventions.

Feature	Epidural steroid injection	Percutaneous discectomy	Spinal ablation (adjunct)
Mechanism	Corticosteroid injection into epidural space to reduce inflammation	Removal of herniated disc material using image-guided instruments	Ablation of sinuvertebral nerve
Technique	C-arm fluoroscopy; transforaminal approach	Image-guided; endoscopic visualization	Performed in combination with discectomy
Anesthesia	Not specified	No general anesthesia required	Not specified
Clinical outcomes	Improvement in PROMIS pain interference scores at 3, 6, and 12 months	Reduced blood loss, shorter hospitalization, smaller incisions, lower inflammatory markers vs open discectomy	Lower pain and disability scores when combined with discectomy vs discectomy alone
Additional notes	Used in patients with cervical radicular pain refractory to conservative therapy	Compared with open discectomy	Adjunctive therapy

PROMIS: Patient-Reported Outcomes Measurement Information System.

ADVANCES IN IMAGE-GUIDED THERAPEUTIC STRATEGIES IN NEUROLOGICAL DISORDERS

Focused ultrasound for essential tremor and Parkinson’s

Magnetic resonance-guided high-intensity focused ultrasound (MR-gHIFU) is a noninvasive technique that ablates targeted intracranial structures without incisions or ionizing radiation. Parkinson’s disease and essential tremor are common movement disorders affecting millions of individuals worldwide. Although both conditions present with tremor, Parkinson’s disease is typically characterized by resting tremor that diminishes with voluntary movement, whereas essential tremor manifests as an action tremor during purposeful tasks such as writing or eating. In addition to pharmacologic therapy, surgical interventions have long been used to manage refractory tremors. In the 21^st century, MR-gHIFU has emerged as a novel treatment modality. This technique uses multiple ultrasound beams to concentrate acoustic energy at a stereotactically defined intracranial target, producing biological effects that depend on the delivered thermal dose; higher temperatures result in tissue ablation[64]. Its noninvasive nature and ability to assess clinical effects during treatment have made MR-gHIFU an appealing option for tremor management.

MR-gHIFU has demonstrated efficacy in patients with medication-refractory tremors. A pilot study evaluating unilateral ventral intermediate nucleus ablation for essential tremor reported significant symptom improvement, with tremor severity decreasing from a mean baseline score of 20.4 to 4.3 at three months and 5.2 at twelve months[65]. Physical performance scores also improved substantially. Similarly, MR-gHIFU thalamotomy has shown benefit in Parkinson’s disease. Schlesinger et al[66] reported immediate tremor resolution in all treated patients, with Unified Parkinson’s Disease Rating Scale scores decreasing from 37.4 to 18.8 one week after treatment. These findings highlight the ability of MR-gHIFU to produce rapid symptom improvement often within days to weeks of the procedure.

In 2018, the FDA approved unilateral MR-gHIFU for Parkinsonian tremor, and in 2025, approval was expanded to include bilateral treatment for both essential tremor and Parkinson’s disease–related tremor. MR-gHIFU is frequently compared with deep-brain stimulation (DBS), which involves surgical implantation of electrodes. Compared with DBS, MR-gHIFU eliminates hardware-related complications and reduces long-term maintenance requirements[67]. However, given its relative novelty and potential for irreversible adverse effects, long-term safety and durability continue to be evaluated, and DBS remains the standard surgical option for many patients. Key innovations in 21^st-century neuroradiology and their clinical applications and outcomes are shown in Table 4.

Table 4 Key innovations in 21^st-century neuroradiology and their clinical applications and outcomes.

Advances in specific areas	Advancements details	Outcomes
Management of cerebrovascular accidents
Stoke triage and imaging analysis	Enables timely and accurate patient evaluation, AI-enabled technology can help reduce prehospital delays, and can rapidly detect large vessel occlusions	Wrist-worn accelerometers integrated with AI algorithms detected stroke as short as 15 mins after onset with accuracy from 0.893 to 0.947[8]. Rapid CTA correctly identified 93% occlusions with a 0.94 sensitivity and NPV of 0.98[10]
Mechanical thrombectomy for acute ischemic stroke	Associated with higher rates of reperfusion, faster neurological improvement, superior functional outcomes	> 90% reperfusion: 89% of patients who underwent mechanical thrombectomy achieve successful reperfusion (> 90% reperfusion) compared to those who underwent alteplase alone (89% vs 34%, P < 0.001)[14]. Median days spent at home within 90 days of discharge were 73 (mechanical thrombectomy) vs 15 (alteplase alone); P = 0.001[14]. Mechanical thrombectomy showed a median gain of 9.3 disability- adjusted life years vs 4.9 in control group receiving alteplase alone[14]
Intracranial-endovascular techniques
Endovascular stenting for intracranial atherosclerotic disease	Prevents plaque protrusion, and reduces rates of major adverse events	“Patients who had carotid stenting procedure using a single-layer carotid stent had statistically significantly more periprocedural neurological complications 8.3% (n = 35) than the double-mesh stent group 2% (n = 3), mostly due to more transient ischemic attacks in the single-layer stent group 4% (n = 17) compared to the double-mesh group 0.7% (n = 1)”[20]
Endovascular aneurysm treatment	Alters aneurysmal hemodynamics to promote progressive occlusion and parent vessel remodeling minimally invasively	Complete occlusion in 55.1% of aneurysm treated with stent assisted coiling (28% recurrence rate) vs 86.7% of aneurysms treated with flow diverters (2.2% recurrence rate)[27]
Endovascular treatment of arteriovenous malformations	Refined microcatheters and newer embolic agents improve safety and efficacy of endovascular treatment	Embolization with Onyx leads to less surgery than embolization with n-BCA (P = 0.0015)[31]
Neurovascular robotics and remote intervention	Can help overcome geographic barriers and expedite stroke treatment	Remote clinicians were able to successfully navigate catheters and guidewires from femoral artery to middle cerebral artery within 15 minutes[36]
Management of brain tumors
Minimally invasive tumor ablation	Suited for small lesions that are difficult to access surgically and help preserve surrounding healthy tissue	Median survival in patients with recurrent glioblastoma increased from 90-150 days to 361 days following LITT[39]
Brain tumor segmentation and treatment planning	Can detect small lesions that could have been overlooked and reduce the risk of overtreatment	A three-dimensional U-Net convolutional neural network for detecting brain lesions on 18F-FET PET imaging achieved an accuracy of 0.9868 during training and 0.9856 during validation, with 100% sensitivity and specificity and no false positives[44]
Advanced perfusion and functional imaging integration	Enables precise localization and preserves eloquent cortex	High-grade gliomas had rCBF of 3.32 ± 1.87 and low-grade gliomas had rCBF of 1.16 ± 0.38. Knowing the CBF can allow precise localization and guide surgeons in resection[49]
Spine interventions
Spine cement augmentation	Percutaneous injection of PMMA under imaging guidance can relieve pain due to osteoporosis	68% of 38 patients with 70 osteoporotic fractures who underwent vertebroplasty experienced complete pain relief within 48 hours[55]. Kyphoplasty showed increased pain reduction vs conservative treatment at 1 month (MD: 2.32, -3.65 to -0.99, P < 0.001)[56]
Minimally invasive spine interventions	Improves targeting accuracy and is associated with reduced blood loss	Patients with cervical radiculopathy who had a transforaminal epidural steroid injection had PROMIS PI at 3-, 6-, and 12-months follow-up that statistically improved by 2.2 (95%CI: 2.1-2.4, P = 0.02), 2.3 (95%CI: 2.1-2.5, P = 0.03), and 2.7 (95%CI: 2.5-3.0, P = 0.03) points, respectively[58]
Image-guided therapeutic strategies in neurological disorders
Focused ultrasound for essential tremor and Parkinson’s	Noninvasive and can assess effects during treatment to monitor ablation	Decreased essential tremor severity from a mean baseline score of 20.4 to 4.3 at three months and 5.2 at twelve months[62]. Unified Parkinson’s Disease Rating Scale scores decreasing from 37.4 to 18.8 one week after treatment[63]
Image-guided gene and cellular therapies	Enables precise visualization of drug delivery and is a potential treatment for neurodegenerative diseases	In a study where three cohorts with Parkinson’s disease received VY-AADC01 gene therapy delivered to the putamen under MRI guidance with doses of ≤ 7.5 × 1011, ≤ 1.5 × 1012, and ≤ 4.7 × 1012 vector genomes, with corresponding putaminal coverage of 21%, 34%, and 42%, PET imaging demonstrated dose-dependent increases in L-amino acid decarboxylase expression (13%, 56%, and 79%), along with reduced antiparkinsonian medication use at six months and improved clinical outcomes and quality of life at twelve months[66]
Radiomics and predictive analytics in neuroradiology	Reveal patterns imperceptible to the naked eye and allows clinicians to anticipate symptom recurrence	In a study evaluating radiomics combined with machine learning to predict the response of metastatic brain tumors to Gamma Knife radiosurgery, the radiomics-based model achieved accuracies of 78% and 87% and sensitivities of 78% and 87%, respectively, compared with 44% and 54% for visual assessment alone[24]

AI: Artificial intelligence; NPV: Negative predictive value; CAT: Computer tomography angiography; CBF: Cerebral blood flow; PROMIS-PI: Patient-Reported Outcomes Measurement Information System-pain interference; LITT: Laser interstitial thermal therapy; n-BCA: N-butyl cyanoacrylate; FET PET: Fluoro-ethyl-L-tyrosine positron emission tomography.

Image-guided gene and cellular therapies (emerging)

Neurodegenerative diseases are difficult to treat because many medications struggle to cross the blood-brain barrier, and when they do, they often affect the entire brain rather than the specific region involved. MRI-guided catheters allow direct delivery of therapeutic agents into the brain, offering several advantages, including more effective bypass of the blood–brain barrier, reduced systemic side effects, and higher drug concentrations at the target site[68]. These systems also enable precise visualization of drug delivery, minimizing the risk of misplacement and injury to surrounding tissue. Improvements in precision delivery have opened the door for gene therapy and stem cell infusion as potential treatments for neurodegenerative disorders, enabling better targeting and improved drug distribution. Technical limitations remain, including shortages of MRI-compatible catheters, low imaging resolution, and the inability to track catheter tips accurately.

Recent advancements have demonstrated the feasibility of MRI-guided catheter delivery for targeted gene therapy in conditions such as Alzheimer’s disease, Parkinson’s disease, and amyotrophic lateral sclerosis. In a study of fifteen patients with advanced Parkinson’s disease and refractory motor fluctuations, participants were divided into three cohorts of five and received VY-AADC01 gene therapy delivered to the putamen under MRI guidance, along with gadoteridol and an adeno-associated virus serotype 2 vector[69]. Cohort doses were ≤ 7.5 × 10¹¹, ≤ 1.5 × 10¹², and ≤ 4.7 × 10¹² vector genomes, with corresponding putaminal coverage of 21%, 34%, and 42%. PET imaging demonstrated dose-dependent increases in L-amino acid decarboxylase expression (13%, 56%, and 79%), along with reduced antiparkinsonian medication use at six months and improved clinical outcomes and quality of life at twelve months. Approximately 95% of patients were discharged within two days. These findings suggest that MRI-guided gene therapy delivery is safe and clinically promising. Even so, clinical translation is constrained by the small sample size and uncertainty in its long-term efficacy and durability. Further studies are needed to evaluate sustained clinical effects and determine its broader application across neurodegenerative diseases.

Radiomics and predictive analytics in neuroradiology

Medical imaging has long been essential in medicine, enabling noninvasive visualization of internal anatomy for diagnosis, planning, and treatment. However, traditional imaging interpretation is limited in its ability to extract all clinically relevant information. Radiomics has expanded imaging from primarily a qualitative to a quantitative tool by using algorithms to extract features such as texture and shape from MRI and CT scans. These features can reveal patterns imperceptible to the naked eye and can be incorporated into models that predict tumor behavior, treatment response, and prognosis. Radiomics “addresses the limitations of subjective radiological interpretation, which can be influenced by interobserver variability and the inherent complexity of neurological structures”[70], promoting more objective, data-driven analysis and advancing precision medicine.

Radiomic signatures have been applied extensively to neurological diseases, particularly brain tumors. Extracted features can be analyzed for correlations with tumor genetics and grade, as well as for predicting recurrence and therapeutic response[71]. In a study evaluating radiomics combined with machine learning to predict the response of metastatic brain tumors to Gamma Knife radiosurgery, the radiomics-based model achieved accuracies of 78% and 87% and sensitivities of 78% and 87%, respectively, compared with 44% and 54% for visual assessment alone[72], highlighting the ability of radiomics to improve prediction of treatment response com-pared with conventional methods.

Radiomics has also shown potential in movement disorders. Innocenzi et al[73] demonstrated that texture-derived gray-level co-occurrence matrix radiomic features extracted from MRI-guided focused ultrasound lesions 24 hours after treatment could predict tremor recurrence at 12 months. This early predictive capability may guide postprocedural monitoring and allow clinicians to anticipate symptom recurrence. While radiomics shows significant promise in neurology, its clinical implementation is challenged by variability in imaging protocols, lack of standardization in feature extraction, and limited external validation, which affect the reliability of radiomic models across clinical settings. Establishing standardized guidelines across institutions can help ensure more consistent interpretation, while incorporating data from multiple institutions for model training may improve generalizability.

Limitations

This narrative review is primarily intended to provide an educational outline for the medical community regarding the advancements in neurodiagnostic in the 21^st century. Given the comprehensive and the narrative nature of the review, it was not feasible to broadly address all possible limitations across the literature. Instead, key limitations relevant to specific studies have been highlighted within the respective sections where they are discussed.

Across the included literature, several studies were limited by short follow-up durations, restricted patient populations, and methodological constraints, including non-randomized designs, small sample sizes, single-center settings, and early trial termination. These factors may introduce selection bias and limit generalizability. Similarly, studies such as those evaluating wearable technologies for stroke detection often lack precise data regarding the exact time of stroke onset and are frequently conducted in hospital settings rather than real-world environments, which may further affect the interpretation and applicability of findings[8]. In addition, cost, infrastructure, and workflow integration challenges present significant barriers, particularly in community hospitals, where implementing advanced technologies such as AI requires trained personnel, system upgrades, staff training, and integration into existing electronic health record systems. Under these constraints, institutions may prioritize immediate clinical needs over adopting newer technologies.

Furthermore, although studies demonstrate strong performance of the deep learning models in controlled research environments, their translation into routine clinical practice remains limited. Barriers include limited external validation across diverse patient populations, variability in protocols between institutions, lack of standardized regulatory frameworks, and concerns pertaining to model interpretability across different health care systems and settings. In addition, many AI models are developed in environmental conditions that do not accurately reflect the complexity and variability encountered in real-world clinical settings. These limitations may hinder broader clinical adoption[74].

AI usage in diagnosis or management of different neurological conditions also raises common concerns, including ethical, legal, and social considerations. Ethical concerns, including patient privacy and the need for transparency and accountability in AI systems, are essential to sustain patient trust. Legal challenges involve liability for AI-related errors and medical malpractice. Social implications, such as potential effects on the patient-physician relationship, patient autonomy, and healthcare disparities, also warrant careful consideration[59]. Furthermore, heterogeneity in long-term outcomes across studies underscores persistent uncertainty regarding the durability and sustained effectiveness of certain interventions.

CONCLUSION

Innovations in the 21^st century have significantly advanced the field of neuroradiology. From minimally invasive techniques to advanced imaging technologies, these developments have transformed clinicians’ approaches to neurological disease and improved patient outcomes. For example, mechanical thrombectomy has improved survival in patients presenting with large-vessel occlusion stroke, while flow diverters and modern stents enable aneurysms to be treated minimally invasively, expanding the range of patients who can be treated. DTI and fMRI enable presurgical planning to maximize tumor resection while preserving critical pathways. The integration of AI into patient care has demonstrated the potential to optimize triage processes by reducing the door-to-treatment time, enhancing diagnostic accuracy, and minimizing errors that may lead to complications, and its pattern recognition abilities may exceed those of human interpretation in certain contexts. Similarly, advances in image-guided interventions have expanded therapeutic options, reduced procedural risks, and improved functional recovery across a wide range of neurological and neurosurgical conditions.

Despite these advances, limitations related to generalizability, study design, and implementation barriers continue to affect the widespread adoption of these technologies. Nonetheless, their demonstrated effectiveness and ability to improve long-term outcomes underscore their growing clinical relevance. Continued refinement of minimally invasive techniques, AI-driven tools, and advanced imaging modalities will further enhance safety, efficiency, and access to care. In addition, future efforts should focus on addressing the barriers to implementation by validating these approaches across diverse clinical settings, increasing dataset size, developing models that can be integrated into a wider range of clinical environments, and reducing implementation costs. As neuroradiology continues to evolve across diagnosis, intervention, and treatment planning, it will serve as a paradigm for innovation in other areas of medicine.