Imaging in Interventional Radiology: 2043 and Beyond

Abstract

Since its inception in the early 20th century, interventional radiology (IR) has evolved tremendously and is now a distinct clinical discipline with its own training pathway. The arsenal of modalities at work in IR includes x-ray radiography and fluoroscopy, CT, MRI, US, and molecular and multimodality imaging within hybrid interventional environments. This article briefly reviews the major developments in imaging technology in IR over the past century, summarizes technologies now representative of the standard of care, and reflects on emerging advances in imaging technology that could shape the field in the century ahead. The role of emergent imaging technologies in enabling high-precision interventions is also briefly reviewed, including image-guided ablative therapies.

© RSNA, 2023

See also the review “Interventional Oncology: 2043 and Beyond” by Elsayed and Solomon in this issue.

Summary

Major developments in interventional radiology over the past century have led to emerging technology that can enable high-precision interventions, including image-guided ablative therapies.

Essentials

• ■ Interventional radiology (IR) will see evolutionary advances in imaging technology over the next 2 decades that will enable new interventional approaches requiring high-precision guidance and quantitative evaluation.

• ■ Advances in image analysis for IR will enable deformable registration of multimodality images, automatic segmentation of structures of interest, quantitation of image features, and a valuable basis for patient-specific modeling and outcomes prediction.

• ■ Artificial intelligence—including machine learning and deep learning approaches—will be integral to advances in imaging technology, image acquisition techniques, image analysis, and predictive modeling.

Introduction

A century of innovation, evolution, and revolution in medical imaging technologies has helped expand the radiologist's role from the diagnostic reading room to the interventional theater. A spectrum of neurologic, thoracic, abdominal, pelvic, and extremities disease is now commonly addressed within the interventional radiology (IR) domain—thanks in part to an arsenal of imaging technologies adapted to the environment, workflow, and imaging tasks in IR. This article briefly reviews the development of such technologies, contemplates their continued evolution in decades ahead, and reflects on the advances in IR capabilities they could enable.

Vignette: The Radiologist and Patient Experience in 2023

It is the morning of January 5, 2023. Dr A arrives for work at the radiology department of her hospital. This morning, she is assigned to CT-guided biopsies, and a quick glance at the procedure schedule reveals the first appointment to be for a biopsy of a 6-mm left upper lobe nodule in a patient with a history of melanoma. The requesting physician would like to determine whether this lesion is a melanoma metastasis versus a primary lung malignant neoplasm or a benign nodule. In the preprocedure area, Dr A discusses the steps of the biopsy procedure as well as the potential risks with the patient. The patient asks her the three most common questions: (a) “When will I know the results?”, (b) “Who will contact me?”, and (c) “Why can't you take the whole nodule out while you're ‘in there’”? She responds to these questions the same way that she has since she was a resident: (a) “Typical results take 3–5 business days and sometimes longer if additional testing on the tissue is required”; (b) “Your primary physician will contact you with the results”; and (c) “If I could, then I would!”

The patient is then wheeled into the CT scanner room. To avoid crossing a fissure, Dr A mentally plans out an off-axial caudocranial trajectory. As she advances her biopsy needle, she calls out to the technologist to scroll through the images on the screen inside the procedure room. “Up!” she yells, and the technologist increments the image to the next cranial section. “Down two!” is the request that follows shortly thereafter. As the biopsy needle approaches the target lesion, the visualization of the lesion diminishes due to beam hardening. She now requests the technologist to toggle the image back and forth from the preprocedure diagnostic CT scan to the most recently acquired intraprocedural sequence, relying on anatomic landmarks such as nearby pulmonary vessel branching patterns and airway anatomy to convince herself that the needle is on the right trajectory. Postbiopsy scans are acquired to rule out pneumothorax, and the patient is taken to the recovery area. There, Dr A informs the patient that the procedure went well and explains the postprocedure instructions. Unless the patient returns for another intervention, this will likely be the last time Dr A meets this patient.

The Evolution and Emergence of Interventional Radiology

Since its inception in the early 20th century, IR has evolved tremendously and is now a distinct clinical discipline with its own training pathways (1–3). These pathways can vary by country, as seen in the centennial article by Almansour et al (4). Imaging remains integral to IR procedures, and as the compendium of IR procedures has increased over time, so too has the frequency with which IR procedures are being performed on dedicated imaging systems. Likewise, imaging is essential for evaluating outcomes of therapeutic interventions performed at IR. Therapies that often started off as experimental have now matured to standardized techniques with high-quality data and have been incorporated into national guidelines for the management of a range of diseases. Still, the spirit of innovation persists in IR, with clinician inventors remaining key figures in driving the field forward (5). Moreover, imaging remains a key component of an interventional radiologist's daily practice, whereby diagnostic interpretation responsibilities are pari passu with interventional duties. However, the paradigm of interventional radiologists as independent practitioners is becoming increasingly common, with many practicing in office-based laboratories.

As IR has evolved over the decades, so too have the imaging techniques. After the initial discovery of the use of x-rays for imaging, the first peripheral angiogram in a patient was performed by Barney Brooks in 1924 with surgical cutdown of the femoral artery and injection of sodium iodide (6). In 1927, the first cerebral angiography was performed with surgical cutdown of the carotid artery by Egas Moniz (7). Percutaneous aortography was first performed by Reynaldo dos Santos in 1929 through direct translumbar puncture (8), although these images were not of great quality due to the limited rate of contrast agent injection through a small needle. In 1953 Ivan Seldinger, a Swedish radiologist, described his technique of accessing a vessel with a thin-walled needle and placing a guidewire into the vessel so that a catheter could then be placed into the arteries (9). With this technique, angiography entered the modern era wherein all vessels of the human body could be accessed with a single puncture site. Endovascular intervention, which used this technique, was first described by Charles Dotter and Melvin Judkins in 1964 (10). In 1974, just a few years after it was made commercially available, CT began to be used for interventional procedures (11).

The State and Future of Imaging Technologies in IR

X-ray Imaging in IR

The current standard of care in IR for x-ray–based imaging relies on two technology platforms: fixed-room C-arms in single- and biplane configurations for two- (2D), three- (3D), and four-dimensional (4D) imaging and helical CT scanners that have been increasingly integrated with fixed-room C-arms in hybrid room configurations. Since the first reported use of stereotaxic CT performed with use of a head frame in 1976 (12), CT technology has majorly advanced. Fixed-room C-arms now universally incorporate flat-panel digital x-ray detectors, offer higher-speed gantry motion for 3D imaging (<5 seconds), and include robotic platforms with a wide range of motion capabilities. Over the same period, helical CT scanners emerged from the “Slice Wars” to break the single-section paradigm with detectors that have multisection capabilities (eg, 64–320 detector rows) with faster scan speeds and volumetric coverage. Scanner designs dedicated to the IR suite now feature longitudinal (z-axis) coverage without table motion thanks to floor rails, forming the basis for the hybrid “angio-CT” suite featuring a C-arm (single-plane or biplane, typically the former) integrated with a CT scanner. These technologies enable a spectrum of imaging capabilities for interventional guidance. Fixed-room C-arms feature 2D fluoroscopy, 2D and 3D digital subtraction angiography, and cone-beam CT image modes as well as emerging 4D digital subtraction angiography and perfusion imaging capabilities. As shown in Figures 1 and 2, real-time 3D-2D registration permits visualization of volumetric data overlaid on real-time fluoroscopy. Helical CT scanners in the interventional suite offer diagnostic-quality CT, perfusion, and dual-energy imaging capabilities, with spectral imaging now emerging with photon-counting detectors.

Figure 1:

Volumetric fusion image of MRI, CT, and cone-beam CT digital subtraction angiography overlaid on real-time fluoroscopy for interventional guidance shows renal cell metastasis in the T10 vertebra.

Figure 2:

Three-dimensional–two-dimensional registration and overlay of a tumor (L3/L4 renal cell tumor) segmented from MRI, registered via cone-beam CT, and overlaid on real-time fluoroscopy. Prior kyphoplasty is also evident.

We anticipate evolution and revolution in these technologies in the decades ahead. For C-arms, improved speed and functionality will be gained via greater degrees of freedom (eg, long-length volumetric imaging), biplanes with 3D imaging from either plane, and novel x-ray sources (eg, new variations on carbon nanotube sources) that enable compact, multisource arrangements. Meanwhile, interventional CT scanners will see new capabilities that parallel those in diagnostic CT with photon-counting detectors, including fast readout, low (zero) electronic noise, high spatial resolution, and spectral imaging capability.

In addition to these advances in imaging hardware, algorithms for improved 3D image quality will be marked by advances in artificial intelligence (AI) that promise to overcome conventional image quality confounders (13), such as quantum noise, metal artifact, and motion artifact. From imaging science, ethics, and regulatory perspectives, crucial questions on the accuracy, reliability, and bias associated with AI algorithms must be addressed and will touch many areas of IR in the decades ahead. Examples include deep learning algorithms to improve image quality, reduce or remove artifacts, automatically detect and segment structures of interest, and extract high-level features from image data to feed predictive models of treatment outcome. In recent years, the use AI tools for diagnostic radiology has expanded—for example, characterization of stroke (14) or cardiac disease (15)—and the potential for such tools in IR is equally exciting.

Cutting-edge areas of AI research include generative AI to synthesize or transform image informati23-0145on from the domain of one modality to another (eg, CT to MRI), offering new approaches for image quality improvement (eg, synthesizing a high-dose image from a low-dose image) and image registration (eg, multimodality registration). Development of AI for assessing uncertainty in the output, such as detecting errors introduced by (epistemic) limitations in the training data or (aleatoric) limitations in the quality of the input data, is also expanding. As AI becomes more commonly used, it will be essential to detect and minimize bias in training sets; understand how to appropriately apply an AI algorithm developed in a given context or population to a new practice or population; federate the learning process across multiple data sets and institutions; and determine reliable methods for continuously training AI algorithms based on new data and monitoring the performance of updated algorithms.

MRI in IR

While MRI is not a standard imaging modality for most interventional procedures, advances in technology have increased its use and provided image-guided interventions with the benefit of exquisite soft-tissue contrast (16). The primary challenges in MRI-guided interventions include access to the patient while imaging is being performed, ideally without moving the patient; the need for MRI-compatible devices; and image quality challenges associated with low-field-strength (eg, 0.2-T) MRI scanners. The clinical deployment of the 0.5-T double doughnut in the 1990s, which included two 60-cm bores about a 58-cm gap, demonstrated the potential advancements associated with improved access (12); however, widespread adoption was limited due to the reduced image quality, and the field of IR largely returned to a closed-bore system. Improvements over the past decade have allowed increased field strength combined with larger bore diameter and shorter bore length without significantly degrading image quality, enabling intervention while the patient remains in the imaging position (eg, Magnetom Free.Max, Siemens Healthineers).

The need for surface coils is also a challenge for interventional procedures, such as biopsy or thermal ablation, as placing the coil over the anatomy of interest often conflicts with percutaneous insertion. Advances in coil design that permit access between coil array elements have enabled coil positioning that does not conflict with needle and probe access. Additional innovations include MRI-compatible needles and probes such as microwave needles (17), radiofrequency ablation electrodes (18), and lasers (19). While first-generation MRI-compatible devices proved to be less robust than traditional devices, scanner designs developed over the past decade exhibit reduced artifacts (depending on device positioning and the MRI sequence) and increased success in clinical use. In addition to image-guided biopsies and thermal ablations, MRI-guided focused ultrasound has demonstrated major success over the past 2 decades (20) and will continue to expand as innovations in MRI guidance translate into clinical practice (21–23).

Innovation in acquisition protocols (eg, scanning speed and artifact correction) has also improved the use of MRI to guide interventions—for example, rapid pulse sequences for real-time guidance, such as the single-shot turbo spin-echo approach for the temporal resolution of fluoroscopy. Novel reconstruction methods have also reduced needle and probe artifacts, allowing for clinical adoption. These fundamental improvements have enabled the exploration of the unique features of MRI, such as the ability to monitor thermal changes of tissue and real-time ablation propagation.

Looking toward the future, we expect innovation in MRI hardware and software to enable larger bore diameter and shorter bore length, allowing increased access to the patient for image-guided interventions. Flexible coil designs with decreased access constraints will likely also become available, and MRI acquisition sequences will continue to develop, increasing efficiency in acquisition with improved spatial and temporal resolution and better soft-tissue contrast. These MRI advances combined with hardware advances for compatible needle and ablation probes and antennas will further improve the accessibility and benefit of MRI-guided interventional procedures.

A combination of technological advances in MRI scanners and new computational capabilities in MR image reconstruction have led to emerging technologies that will improve MR image guidance capabilities, including mobile MRI. Mobile MRI systems with a small footprint, large inner bore, sealed self-contained gantry, and low power requirements (operating simply from wall power and without venting) will broaden the application of interventional MRI, permitting a single system to service multiple interventional suites. While such systems may operate at field strength considerably less than 1.0 T, advances in image reconstruction with AI algorithms will offer high-speed scanning and sufficient image quality from sparsely sampled, noisy data.

Development of higher-field-strength MRI will continue to increase MRI soft-tissue contrast and spatial resolution, but integration of this for IR may be challenging due to the traditional limitations of MRI described earlier. However, the innovation of both magnet and coil design, combined with developments in image analysis, AI, and augmented and virtual reality, will enable increased opportunities for the use and integration of high-field-strength MRI for interventions.

Advances in AI will greatly impact MRI; for example acquisition and reconstruction strategies will likely use AI for improved tissue contrast, spatial resolution, quantification, and speed and reduced artifacts. Recent advances include real-time reconstruction using deep learning techniques to track the delivery of interventions (24–26). Motion artifacts can also be reduced using deep learning, including identification and removal of phase-encoding lines (27) degraded by motion or reduction of streaking artifacts caused by free breathing (28). Deep learning has also enabled MR fingerprinting for fast, accurate, and reproducible acquisition and reconstruction of quantitative MRI scans (29–31).

US Imaging in IR

Although medical US imaging traces its roots to the 1930s, it was in the mid-1970s that digital scanners became broadly available and overcame the cumbersome, impractical nature of previous systems. Since the turn of the millennium, US imaging has emerged in many medical contexts as a new “stethoscope” via small, low-cost, point-of-care systems that introduced the technology to fields such as emergency medicine and rheumatology. The decades ahead will see even broader use as costs decrease and capabilities increase. In interventional contexts, robot-assisted US imaging will improve positioning accuracy, reproducibility, and registration with other imaging modalities (32). Concomitant advances in AI assistance will reduce operator variability, improve image quality (analogous to the aforementioned improvements for CT and MRI), and extract more quantitative information, including information gleaned by AI algorithms from speckle (33).

One emergent US imaging technology that will change IR practice in the next 20 years is photoacoustic imaging, which opens new possibilities in contrast resolution and quantitative imaging (34). Indeed, conventional US limitations related to tissue depth can be resolved in many interventional contexts by minimally invasive introduction of the photoacoustic system. For example, quantitative measurement of oxygen concentration with use of photoacoustic US will offer the interventionalist a label-free biomarker of hypoxia analogous to blood oxygen level–dependent MRI. Elasticity imaging with US will also advance beyond conventional compressive elastography techniques and present new opportunities for IR by means of shear-wave imaging and access to quantitative measures of tissue biomechanical properties (35). Especially for ablative techniques, this capability will provide a valuable biomarker for therapy. Novel contrast agents will also advance the role of US in IR, including targeted microbubbles, nanobubbles, and nanodroplets capable of extravasation and manipulation of state by means of ultrasonic excitation. The relatively lower regulatory hurdles associated with such agents (compared, for example, with quantum dots) will ease their translation to new clinical scenarios.

The aforementioned advancements will be tied in part to advances in US transducer technologies, including 2D arrays, the burgeoning number of channels that can be incorporated within a single transducer element, and the ability to perform volumetric US from a fast, single sweep with improved in-plane image quality (36). Among the exciting platform technologies for such advances are capacitive micromachined ultrasonic transducers and single-crystal transducers that enable beam forming on small circuits embedded in the transducers. Not only will this enable arrays with higher channel count, but it may also allow for the light source in photoacoustic US to be used in novel (eg, intravascular) contexts.

Molecular Imaging in IR

Following decades of investigation, several advances—including the use of filtered back projection as promulgated by Hounsfield for CT, development of fluorine 18 (¹⁸F) fluorodeoxyglucose as a practical tracer, and (later) dual-detector scintillation cameras with coincidence detection—led to the availability of PET for molecular imaging in medical practice. Molecular imaging enables tissue-specific assessment of biologic processes associated with disease. For instance, fluorodeoxyglucose PET/CT is routinely performed to identify tumors with increased metabolism indicative of cancer. For other metastases, technetium 99m methylene diphosphonate SPECT is used to detect increased bone turnover within bone metastasis. MRI can also be used to detect molecular characteristics, such as diffusion of water molecules with diffusion-weighted imaging and detection of blood oxygen levels with susceptibility-weighted imaging. Novel targeted contrast agents and radiopharmaceuticals created by conjugating antibodies with gadolinium or radioactive isotopes may enable visualization of overexpressed proteins in specific tissues. Currently, the most common use of molecular imaging in IR is for preprocedural planning of biopsies, tumor ablation, and catheter-directed tumor embolization. Image fusion and precise targeting is possible; however, there can be limitations with fusing images obtained in a separate setting (37). Real-time interventional PET/CT can increase the accuracy of molecular image-guided biopsies (38,39) as well as help with guidance and intraprocedural confirmation of tumor ablation based on loss of cancer-related ¹⁸F fluorodeoxyglucose uptake (40–42).

Multimodality Imaging in IR

The modern standard of care uses a variety of multimodality imaging paradigms, primarily with task-specific combinations of US, fluoroscopy, CT (or cone-beam CT), and MRI. Due to advantages of speed, cost, and ease of use, US and fluoroscopy are fairly pervasive, although methods to register these modalities with other image data are still lacking. Hybrid room configurations are evident in two main forms: (a) “CT angio” hybrid configurations of a C-arm (fluoroscopy and cone-beam CT) and CT scanner and (b) “MR angio” hybrid configurations of a C-arm and MRI scanner. Such hybrid configurations involve an ever-increasing degree of system integration to permit fast transition between the imaging systems and registration of image data. As shown in Figures 3 and 4, the ability to fuse multimodality volumetric data provides exquisite visualization of the therapeutic target, feeding vasculature, and adjacent normal anatomy. The utility of such hybrid configurations varies with the interventional task. For instance, fluoroscopy/cone-beam CT is used for guidance or “quick” checks on device localization, whereas CT or MRI is used for high-quality verification of treatment delivery. Alternatively, in some scenarios, the value of such hybrid configurations may be the capability to perform a broad range of procedures within a given room, including procedures that only use one of the imaging technologies.

Figure 3:

Image shows falcotentorial meningioma with MRI tumor segmentation and image fusion with cone-beam CT digital subtraction angiography and cone-beam CT diameter spherical volume for preoperative mapping and assessment of the eloquence of venous structures. Occlusions of the straight sinus and vein of Galen are depicted.

Figure 4:

Image shows recurrent superior falcine meningioma with MRI tumor segmentation and image fusion with cone-beam CT digital subtraction angiography and cone-beam CT diameter spherical volume for preoperative mapping and assessment of the eloquence of venous structures. Occlusions of the sagittal sinus and middle meningeal tumor supply are depicted.

Moving forward, hybrid suites will likely be used for a broader spectrum of IR procedures. Moreover, it will likely be increasingly valuable for medical centers to have hybrid suites that allow for multiple types of procedures to be conducted in a single room, and in the future, distinct roles for these suites in high-volume, centralized hospital settings and lower volume and/or geographically distributed clinical services may become apparent. PET/MRI emerged over the previous decade with potential applications in IR (43)—and PET/CT in the decade before that—and their value to IR in the decades ahead will be proportionate to that of molecular imaging in guiding or assessing interventional procedures (44).

Key to the utility of such multimodality imaging–guided interventions are the aforementioned technological advances (eg, faster scan modes) and increased functional capabilities. As imaging technologies advance, the importance of fully integrating images for planning, treatment delivery, and therapeutic assessment becomes even more evident. Accurate image registration is also essential to harness the benefits of advanced imaging. The development, validation, and clinical deployment of deformable image registration over the past 2 decades has enabled more accurate evaluation of longitudinal changes in tumors and normal tissue, which is essential for treatment assessment, retreatment, and multimodality treatments (45). Increasing the speed of image registration to enable near real-time integration of multimodality imaging for guidance will be an important area of innovation in the immediate future. In addition to registration, the presentation of image fusion is essential for optimizing the use of advanced images and multimodality imaging. Simple advances, such as large in-room monitors, have provided initial levels of integration; however, the emergence of augmented and virtual reality promises to enable further advancements in clinician education as well as for treatment accuracy, precision and efficiency, reduced imaging dose, and reduced rates of complications (46–50).

Imaging Enabling Advanced Interventions

Advances in imaging like those described earlier will enable new interventional approaches, particularly for minimally invasive therapies that require a high degree of accuracy and precision in placing a therapeutic device and/or measuring the distribution of the therapeutic effect. Moreover, advanced imaging technologies and analysis will accelerate the emergence of experimental therapeutics by minimizing the variations in geometric localization, thereby properly attributing variations to the therapeutic action and physiologic response. Interventional imaging will not only provide high-precision guidance; it will also be the basis for extraction of quantitative, image-based features that will feed predictive models of therapeutic outcome. As demonstrated in Figure 5, the ability to deformably register volumetric images throughout the planning, guidance, validation, and follow-up stages of therapy is key for quantitatively and reliably evaluating treatment delivery.

Figure 5:

For liver tumor ablation, multimodality images are segmented using deep learning–based tools and deformably registered to resolve geometric alignment in each phase of planning, treatment, and follow-up. (A) Pretreatment contrast-enhanced CT scan obtained at the start of the thermal ablation procedure with (B) three-dimensional (3D) visualization. Automatically segmented structures include the liver (cyan), tumor (green), and vasculature (purple). (C) Deformable image registration maps the tumor onto a noncontrast CT scan acquired for image guidance, showing the needle in position to ensure targeting accuracy. (D) Visualization of automatically segmented structures according to the color legend in A. Finally, deformable image registration maps the tumor onto posttreatment contrast-enhanced CT scan for visualization relative to the ablation margin (orange). Follow-up (E) MRI and (F) contrast-enhanced CT scans are shown with the liver segmented and tumor mapped with use of deformable image registration. A = anterior, P = posterior. Image series from the Cover-All Study (ClinicalTrials.gov identifier NCT04083378) for liver tumor ablation (courtesy of Bruno Odisio, MD, The University of Texas MD Anderson Cancer Center, principal investigator of the Cover-All Study).

The U.S. Food and Drug Administration approved the use of Hansen Medical's endovascular robot for peripheral interventions in 2012 and Corindus Vascular Robotics’ robot for coronary interventions in 2018. Initially these tools, along with others being developed, focused on reduced radiation exposure to the operator and the ability for a remote operator to perform procedures. However, the capability of a robot may eventually exceed that of a human operator. For instance, neurointerventional software is being developed to program robots to actively maintain microcatheter stability within brain aneurysms and to aid in precise deployment of stents in the brain. Like the development of self-driving cars, neural networks can be developed to learn the nonlinear skills of the Seldinger technique, actively advancing wires into distal locations and then advancing catheters and microcatheters over these wires. Self-driving endovascular robots may one day improve the speed, safety, and accuracy of endovascular procedures.

The potential impact of AI in IR extends far beyond image acquisition and reconstruction (51), as noted in the aforementioned examples. Early research demonstrates the potential of AI to aid in optimal patient selection for treatment and evaluation of treatment outcomes (52). Additionally, AI has been shown to improve the efficiency and accuracy of planning techniques for therapeutic interventions and interventional biopsies by performing segmentation of the tumor, tumor-bearing organ, and sensitive normal tissues, as well as aiding in treatment decision-making (53). Early research has also demonstrated the ability of AI to predict local control by using pre- and postablation images. These studies demonstrate the potential for AI to optimize not only patient selection for thermal ablation techniques, but also the number and location of needles, power, and duration of ablation.

Conclusion

IR has become a cornerstone in health care, encompassing almost all aspects of patient care. What may have once required an open surgical procedure or, more likely, would not have been considered possible is now routinely performed in this specialty. Engagement at the interface of IR and surgical specialties will continue to be vibrant. Over the past 20 years, IR approaches have advanced in areas that were once the sole domain of surgery, including spinal interventions (eg, vertebroplasty), ablative therapies throughout the abdomen and pelvis, and treatment with high-intensity focused ultrasound. Such advances will continue in the 20 years ahead, and numerous opportunities for interdisciplinary or multidisciplinary coordinated care between IR and surgery will emerge, including neurointerventional radiology in coordination with neurosurgery as well as percutaneous fracture fixation and other bone interventions in coordination with orthopedic surgery. Similarly, opportunities for coordinated care between IR and emerging small-molecule, antibody, and cell-based immunotherapies will increase, with IR approaches offering geometrically precise delivery and image-based measurement of the resulting biodistribution and treatment response.

Over the past 20 years, learning-based algorithms using AI have emerged among the most exciting areas of medical imaging research and likely will be essential tools for future clinical practice (51). Radiology was among the medical disciplines best positioned to lead the development of AI techniques, knowledgeably translate algorithms for clinical use, and illuminate many of the challenges in deployment and management of such models in a reliable, unbiased manner. Factors contributing to such an advantageous position include access to a rich volume and variety of image data, expertise suitable to truth annotation, hospital IT systems for computerized data management, and collaboration with computational scientists and medical physicists with clear understanding of clinical data and challenges. The first decade of AI advancement was not without pitfalls (54,55), and the decades ahead will see research and clinical activity that aims to increase the capabilities of AI methods while also illuminating and minimizing underlying bias, improving the understandability and interpretability of model outputs, and facilitating the transfer and continuous learning of algorithms within and between patient populations. Generative AI models (56,57)—including image generators and large language models—appear to be similarly poised for a major impact on interventional medicine. Numerous new and well-established challenges in the training and generalization of deep neural networks must be addressed for such methods to find reliable use in IR, including the need for disease-specific training sets and streamlined methods to facilitate conventionally laborious methods for truth definition. While diagnostic radiology was among the most exciting areas of data-scientific approaches in medicine over the past decade, IR will see commensurate activity moving forward, where advanced AI methods will enable new capabilities in image guidance, quantitative evaluation of therapeutic effect, and predictive modeling of patient-specific outcomes.

The decades ahead will see the continued evolution of IR along themes that arose at the field's nascency as well as the emergence of new priorities enabled by the tremendous progress in imaging and procedural technology (Table). The mantra “cheaper, better, faster” has long been used to encapsulate the capabilities of the minimally invasive interventions that IR is known for, and certainly providing effective, cost-efficient, and safe treatments with decreased side effects will remain the cornerstone of the specialty. However, in addition to this priority, we also anticipate that advances in imaging and robotics will provide the opportunity to pursue new goals, including patient-centered care, timeliness, and shared decision-making. As IR continues to grow as a clinical specialty, the duties of the field will also expand to include such things as running outpatient clinics, cultivating long-term relationships, and participating in multidisciplinary care management discussions. The major limitation for implementing such processes currently is time; however, the time gained by the anticipated increases in efficiency for all aspects of an IR procedural workflow by means of imaging and technology advances will allow for the reallocation of time toward these goals. Ultimately, we expect that such advances will strengthen the most critical component of an IR intervention, namely, the physician-patient relationship.

Summary of Major Advances and Anticipated Clinical Benefits to Be Gained in Interventional Radiology in the Decades Ahead

Vignette: The Radiologist and Patient Experience in 2043

It is the morning of January 5, 2043. Dr A arrives to work at the Department of Interventional Radiology. She greets her first patient, someone she has known since the patient first presented to the hospital and she met along with other specialists in a shared multidisciplinary clinic. The patient has a history of melanoma, with a new 6-mm lung nodule in the left upper lobe identified at recent imaging. Molecular imaging and liquid biopsy results have confirmed that this lesion represents a melanoma metastasis. The goal of today's intervention, therefore, is not histologic diagnosis. Rather, Dr A will collect tissue from the lesion for complete genome analysis as well as profiling of the intratumoral immune cell population. Moreover, additional samples will be acquired to create a bank of tumor-infiltrating T lymphocytes that may be administered systemically as well as to generate a patient-specific murine model to test alternative systemic therapies. At the same time, she will eradicate the lesion with percutaneous ablation.

Based on the patterns of imaging features present on preprocedure images, she elects to proceed with cryoablation, as the AI-driven models based on thousands of previously treated patients predict this modality to have the best local tumor control rate as well as potential for local immune stimulation and lowest chance for complications. Since Dr A has already discussed the procedure and its rationale with the patient previously in a dedicated clinic visit, this morning's conversation can be centered around the patient's recent vacation. The patient is then wheeled into the state-of-the-art multimodality IR suite (Fig 6). The preprocedure imaging is fused to the procedural imaging, and the target lesion is automatically segmented. Dr A remains in the control room while she confirms the computer-generated percutaneous trajectory that is optimized for parenchymal length, traversing of fissures, and puncture of blood vessels. Once confirmed, she loads the appropriate needle into the interventional robot. The patient undergoes intravenous sedation, and the overlying skin is prepared. The procedure is initiated as Dr A instructs the robot to advance the needle along the prescribed trajectory. Despite the patient's free breathing and the beam hardening artifact from the metallic needle, the visualization of the target lesion remains unimpaired due to advanced AI-driven reconstruction techniques augmented with real-time 4D registration technology. Dr A maintains clear visualization of anatomy and the interventional device in real time with use of an augmented reality display that keeps her attention on the patient with full situational awareness of the interventional suite. Once the requisite biopsy cores are acquired, Dr A advances a cryoablation needle through the biopsy cannula into the target lesion. With use of real-time imaging monitoring and segmentation of the ice ball, the optimal freeze and thaw duration times are predetermined for approval by Dr A. Postprocedure imaging confirms that no complications occurred and that the ablation was successful in completely destroying the lesion. Following the intervention, the patient is taken to the recovery area, where Dr A explains the successful outcome of the procedure. She then reminds the patient of their follow-up clinic appointment in 2 weeks, where they will discuss any necessary next steps in the patient's care.

Figure 6:

A view of the future in interventional radiology. With multimodality image guidance and robotic assistance, the radiologist approaches the target in a streamlined manner with minimal injury to normal tissues. Real-time deformable image registration accounts for tissue changes, visualized by the radiologist with use of an augmented reality (AR) viewer that maintains visual field on the patient and situational awareness throughout the interventional suite. Targeted therapy is delivered under image guidance with real-time, quantitative feedback measuring the distribution of therapeutic effect and validation of treatment delivery.