Quantitative Ablation Confirmation Methods in Percutaneous Thermal Ablation of Malignant Liver Tumors: Technical Insights, Clinical Evidence, and Future Outlook

Abstract

Percutaneous image-guided thermal ablation is an established local curative-intent treatment technique for the treatment of primary and secondary malignant liver tumors. Whereas margin assessment after surgical resection can be accomplished with microscopic examination of the resected specimen, margin assessment after percutaneous thermal ablation relies on cross-sectional imaging. The critical measure of technical success is the minimal ablative margin (MAM), defined as the minimum distance between the tumor and the edge of the ablation zone. Traditionally, the MAM has been assessed qualitatively using anatomic landmarks, which has suboptimal accuracy and reproducibility and is prone to operator bias. Consequently, specialized software-based methods have been developed to standardize and automate MAM quantification. In this review, the authors discuss the technical components of such methods, including image acquisition, segmentation, registration, and MAM computation, define the sources of measurement error, describe available software solutions in terms of image processing techniques and modes of integration, and outline the current clinical evidence, which strongly supports the use of such dedicated software. Finally, the authors discuss current logistical and financial barriers to widespread use of ablation confirmation methods as well as potential solutions.

Keywords: Ablation Techniques, CT, Image Postprocessing, Liver

Supplemental material is available for this article.

© RSNA, 2025

Summary

Intraprocedural use of ablation confirmation software is critical and allows for quantitative measurement of the minimal ablative margin after percutaneous thermal ablation of liver tumors and immediate re-ablation of insufficient margins.

Essentials

• ■ The minimal ablative margin (MAM) has been identified in multiple independent studies as the main predictor for local tumor progression after percutaneous thermal ablation of several types of primary and secondary malignant liver tumors.

• ■ Over the past few years, a large variety of software packages with different technical components have been developed and evaluated for quantitative assessment of the MAM.

• ■ Ablation confirmation software usually relies on fast and accurate segmentation of the tumor, ablation zones, and the liver if required by the registration algorithm.

• ■ As well as registration of pre- and postablation images, and both segmentation and registration, have been significantly improved with artificial intelligence–based image processing techniques.

• ■ The current evidence involving ablation confirmation software is limited to retrospective studies of common types of malignant liver tumors, such as hepatocellular carcinoma and colorectal liver metastases.

• ■ It is expected that ongoing prospective trials are set to provide the required level of evidence to support ablation confirmation as an integral part of standard-of-care for patients undergoing percutaneous thermal ablation.

Introduction

Percutaneous thermal ablation is a widely used minimally invasive, curative-intent local treatment for selected patients with primary and secondary malignant liver tumors. The efficacy of thermal ablation, whether used in isolation or in conjunction with systemic therapies, has been demonstrated in multiple large retrospective series, meta-analyses, and prospective studies (1,2) and is listed in several treatment guidelines as one of the cornerstones of local treatment (3–9).

Although thermal ablation has inherent advantages over surgical resection due to its minimally invasive nature, its efficacy with respect to local tumor control has varied widely, ranging from 5% to 51% (10,11). This variation in local efficacy has primarily been attributed to the inability to obtain tissue samples around the entire ablation volume for histologic confirmation of complete tumor ablation with clear margins, thereby limiting local end point assessment of thermal ablation. Although confirmation of tumor-free margins (R0 resection) during surgical procedures can be achieved through intraoperative microscopic examination of frozen tissue slices, confirmation of complete tumor coverage during thermal ablation is achieved indirectly through noninvasive macroscopic imaging techniques, such as CT (11).

In an effort to overcome this constraint, reporting guidelines published in the early 2010s suggested assessing the technical success of thermal ablation based on whether the tumor, plus an additional isotropic ablative margin of 5 to 10 mm, has been entirely enveloped by the ablation volume (12). Historically, this assessment involved visually inspecting pre- and postablation images placed side-by-side (13). However, a study revealed that such visual inspection methods can lead to misjudgment, even among experienced interventional radiologists (14). Therefore, it is now recommended that the technical success of thermal ablation be confirmed through ablation confirmation techniques employing rigid or nonrigid image fusion (11,15,16). Image fusion, which overlays and aligns the pre- and postablation images, represents an improvement over the side-by-side comparison of coregistered images. However, image fusion for qualitative, and therefore subjective, evaluation is still susceptible to interoperator variability. Consequently, several software packages for ablation confirmation have been developed to quantitatively measure the minimal ablative margin (MAM).

In this review, we provide an overview of the terminology and image processing techniques involved in MAM measurement, as well as the current clinical evidence regarding the optimum MAM and its prognostic value. Current logistical and financial barriers to the widespread use of ablation confirmation methods will also be discussed, along with potential solutions to address these constraints.

Terminology

The MAM is defined as the shortest distance between the tumor and the edge of the ablation zone, as observed on cross-sectional images. Given that the coagulative necrosis from the thermal ablation process effectively eliminates the tumor depiction on these images, the tumor contour is derived from a preablation CT image. In contrast, the ablation volume contour is derived from a postablation CT image. The preablation image can be the last diagnostic image obtained before the procedure or, ideally, an image obtained intraprocedurally before the placement of ablation probes or the delivery of ablation energy. The postablation image can be either the final image obtained during the procedure (preferably) or the initial follow-up image.

Technical success in tumor ablation assesses whether the tumor was treated according to protocol and fully covered by the ablation zone, without specifying the attainment of optimal ablation margins (12). With the introduction of ablation confirmation techniques, the technical success of thermal ablation can now be stratified based on the achieved MAM and its associations with local oncologic outcomes. Analogous to the surgical concept of R0 resection (microscopically negative tumor), A0 ablation defines the MAM threshold linked to optimal local tumor control for a specific tumor histologic subtype and ablation confirmation methodology (17). A1 ablation denotes complete tumor coverage with inadequate margins, and A2 ablation signifies incomplete tumor coverage, in which part of the ablated tumor remains outside the ablation zone. It is important to note that, due to the limited number of prospective studies on the use of ablation confirmation software, most of the current A0 definitions are derived from retrospective studies of patients treated with percutaneous thermal ablation without applying intraoperative ablation confirmation methods (17–21). The MAM can be assessed either with intraprocedural imaging (ie, images acquired during the ablation procedure) or periprocedural imaging (ie, images acquired before or after the procedure, typically on a different day).

When describing clinical ablation outcomes, residual unablated tumor refers to the presence of tumor foci at the ablation margin observed on the first follow-up image, and local tumor progression refers to the reappearance of tumor foci in contact with the ablation zone after at least one contrast-enhanced follow-up image has documented adequate ablation (12,15).

Technical Components of Software-based MAM Computation

The workflow for ablation confirmation software includes cross-sectional image acquisition (before and after ablation), segmentation of the tumor, ablation volume, liver surface, or its vasculature (if required by the algorithm), registration of pre- and postablation cross-sectional images, and computation of the MAM. A conceptual overview is provided in Figure 1, and a representative case is illustrated in Figure 2. The technical components of these workflow steps are discussed below, and pros and cons are outlined in Table 1.

Figure 1:

Illustration of ablation confirmation software workflow shows cross-sectional image acquisition (before and after ablation); segmentation of tumor (green), ablation volume (yellow), and liver contours (blue and red, if required by algorithm); registration of pre- and postablation cross-sectional images; and display of required 5-mm ablative margin (orange) and computation of the 2-mm minimal ablative margin (MAM).

Figure 2:

CT images in a 70-year-old man with a liver metastasis in segment VIII from a neuroendocrine tumor who underwent percutaneous thermal ablation of one lesion as part of the COVER-ALL (Clinical Impact of a Volumetric Image Method for Confirming Tumor Coverage With Ablation on Patients With Malignant Liver Lesions) trial (66). (A) The 3-cm tumor and the liver were segmented (green) at the intraprocedure preablation contrast-enhanced CT (CECT) and the intended 5-mm minimal ablative margin (MAM) visualized (yellow). (B) A microwave ablation probe was placed, and accurate placement was confirmed at noncontrast CT. The tumor (white arrowhead) is outlined in green and the intended 5-mm MAM in yellow. (C) Following the ablation, deformable registration of the postablation CECT image showed that the tumor (white arrowhead) was not completely covered, and immediate reablation was performed. (D) After reablation there was still a small area of uncovered tumor (white arrowhead), which was again reablated. (E) Then, repeat reablation and artificial intelligence–based segmentation of the ablation zone (orange) was performed, and the final MAM was computed at 4.3 mm. Additional reablation was not pursued at this point because the volume of insufficient coverage was 0.05 mL. (F) At 2-year follow-up imaging, no local tumor progression was observed on CECT scan.

Table 1:

Overview of Pros and Cons of Different Technical Aspects of MAM Assessment Methods

Pre- and Postablation Image Acquisition

CT Imaging

Today, CT is widely recognized as the preferred imaging method for planning and guiding liver ablation procedures, with the routine use of contrast medium to aid in locating tumors and visualizing the ablation zones (10,11). Intraprocedural contrast-enhanced CT (CECT) for confirming ablation results requires the use of iodine-based contrast agents, which carry the theoretical risk of contrast material–associated acute kidney injury. A typical multiphase CT requires a dose of approximately 100–120 mL of intravenous contrast medium (22,23). Therefore, a patient should be able to tolerate at least 200 mL of contrast medium to allow for one CECT scan before and at least one after ablation. A repeat CECT for reconfirmation after reablation might result in reaching the maximum allowable contrast material dose (5 mL × body weight [in kilograms] ÷ serum creatinine [in milligrams per deciliter], with a maximum dose of 300 mL for a healthy adult) (24).

An alternative intraprocedural contrast-enhanced imaging modality is CT during hepatic arteriography (CTHA), in which contrast medium is directly injected into the common or proper hepatic artery via an angiographic catheter. This approach typically uses a contrast medium dose of less than 20 mL per CT scan (25,26). Due to the low dose, multiple contrast-enhanced images can be acquired throughout the procedure without reaching the maximum allowed dose, as demonstrated in Figure 3.

Figure 3:

CT and MR images in a 42-year-old woman with a colorectal liver metastasis in segment VII who underwent percutaneous CT during hepatic arteriography (CTHA)–guided liver ablation as part of the STEREOLAB (Stereotactic Liver Ablation Assisted with Intra-Arterial CT Hepatic Arteriography and Ablation Confirmation Software Assessment) trial (53). (A) Baseline contrast-enhanced CT (CECT) image obtained 1 month before the procedure showed a tumor measuring 2.5 cm (arrowhead). (B) Intraprocedure CTHA image shows a 2.5-cm tumor (green) with ring enhancement segmented in green and the intended 5-mm minimal ablative margin (MAM) in yellow. (C) Subsequently, stereotactic placement of two ablation probes (white asterisk) was done and repeat CTHA imaging was used to confirm their accurate positioning. (D) After 10 minutes of ablation, ablation confirmation (AC) using a deformable image registration of the postablation CTHA image showed insufficient margin on the superior and lateral aspect of the tumor (arrowhead). (E) Repeat ablation was performed immediately followed by more CTHA scan. AC using a biomechanical deformable image registration–based MAM confirmation showed sufficient margins (MAM > 5 mm). The total contrast material dose was 100 mL despite acquisition of multiple images with contrast enhancement. (F) MR image obtained 6 months after the procedure revealed involuting nonenhancing ablation zone on subtraction image, indicating no local tumor progression.

US Imaging

Although US is widely used for image guidance, it has limitations for quantitative assessment of MAM due to gas artifacts during ablation and its two-dimensional nature. To date, approaches using three-dimensional (3D) contrast-enhanced US have been limited to visual confirmations (27,28) or the integration of fusion imaging systems (29) in single-center experiences. However, challenges remain regarding the systematic and standardized use of US for precise MAM evaluation, considering operator dependence, critical and potentially complex registration issues, imprecise auto-registration, and the generally limited available data.

MRI Studies

As with CT imaging, contrast enhancement for MRI allows for accurate tumor and ablation zone depiction (11). Contrast-enhanced MRI is paramount for diagnosing and monitoring liver lesions, especially in patients eligible for local treatments (30), to precisely depict tumor burden. Due to logistical and financial challenges, MRI is rarely used for intraprocedural image guidance during thermal ablation. Therefore, MRI scans acquired periprocedurally are primarily used for image fusion with intraprocedural CT images for tumor depiction and MAM evaluation.

Intraprocedural versus Periprocedural Timing of Pre- and Postablation Imaging

To date, studies have found that the optimal imaging approach for the ablation procedure itself and for quantifying MAM is intraprocedural imaging with pre- and postablation CECT (22,31). Importantly, MAM quantified using intraprocedural imaging has been shown to be significantly more predictive of local tumor progression than MAM quantified using periprocedural imaging (ie, images acquired before and/or after the day of the ablation procedure) (31,32,66). This discrepancy is mainly due to the shrinkage of the ablation zone and potentially less accurate registration caused by organ deformation and patient positioning. An example of different MAM assessment results is shown in Figure 4.

Figure 4:

CT images in a 41-year-old woman with colorectal liver metastases who underwent percutaneous thermal ablation of two lesions. (A) One lesion in segment VII was measuring 1.2 cm, and one lesion in segment VIII was measuring 0.6 cm. (B) Minimal ablative margin (MAM) assessment at immediate postablation contrast-enhanced CT (CECT) shows sufficient margins of 5 and 3.7 mm with volumes of insufficient coverage of 0 and 0.04 mL, respectively. (C) MAM assessment on the first imaging follow-up 30 days after ablation shows both margins (arrowhead) were 0 mm (segment VIII 0 mm; MAM not shown on this image), and volumes of insufficient coverage were 0.99 and 0.35 mL. (D) At 2-year follow-up, no local tumor progression was observed.

Image Segmentation

All ablation confirmation software uses the tumor and ablation zone contours to calculate the MAM between them. Some software may also require liver or vasculature contouring depending on the registration algorithm or any corrective measures needed for specific cases, such as subcapsular tumors. Manually segmenting the tumor, ablation zones, and liver is extremely time-consuming, heavily reliant on operator input, and therefore prone to bias.

The development of automated segmentation has been facilitated by large datasets and competitions such as the Liver Tumor Segmentation challenge at meetings of the Medical Image Computing and Computer Assisted Intervention Society (34). Although numerous algorithms for automated liver and tumor segmentation have been published, few have achieved clinically acceptable performance. Convolutional neural network architectures like SegNet (35) and U-Net (36) have allowed clinical implementation of some automated but supervised algorithms. Although public algorithms suffice for liver and tumor segmentation (37), ablation zone segmentation still requires specialized models trained on relevant images (38).

Image-to-Image Registration

Pre- and postablation images must be in the same frame of reference for MAM computation, which is achieved by image-to-image registration—a common problem in medical image processing with a wide variety of algorithms and techniques available. Each algorithm operates under a specific set of assumptions and conditions depending on the particular application (39). Fundamentally, image registration involves taking a fixed image (eg, a postablation image) and a moving image (eg, a preablation image), transforming the moving image to optimize a similarity metric. The transformation can be rigid or deformable, and the similarity metric may be based on intensity values or contours. Additionally, biomechanical algorithms incorporate finite element methods, offering inherent benefits in imaging flexibility, a critical factor during liver ablation (see below).

The main challenges in image registration for confirming liver ablation include organ deformation due to breathing, the placement of ablation probes, patient manipulation, and the procedure itself. Furthermore, thermal ablation dehydrates the tissue, and in the case of microwave ablation, considerable tissue shrinkage occurs (40). If the tumor is close to surrounding critical structures, displacement maneuvers such as hydrodissection or gas insufflation can lead to additional deformations. If periprocedural imaging is used instead of intraprocedural imaging, differences in patient positioning and breathing can further complicate image registration.

MAM Computation

MAM computation is typically accomplished using one of two primary methods: distance-based and volume-based. Regardless of the specific measurement method, MAMs are often categorized into groups, such as ≤0 mm, 0 to <5 mm, and ≥5 to <10 mm or ≥10 mm, corresponding to incomplete ablation (A2), complete ablation with insufficient margin (A1), and complete ablation with sufficient margin (A0) (19,20,41,42). However, the specific thresholds may vary between studies, tumor histologic subtypes, and measurement accuracy.

Distance-based Method

In the distance-based method, the shortest distance between the registered tumor and ablation volume contours is determined. Generally, this distance is computed by transforming the tumor segmentation mask into a 3D discrete distance map and identifying the ablation volume contour voxel with the smallest distance value (43). The measurement is performed on the discrete 3D voxel grid. Alternatively, the contours can be converted into 3D models to compute the shortest distance in continuous space (Fig 5, top). With distance-based MAM, a larger value indicates a larger margin, and a value of 0 mm or less indicates that the tumor itself is not completely covered by the ablation zone.

Figure 5:

Illustration of the difference between distance-based (top) and volume-based (bottom) minimal ablative margin (MAM) computation. Distance-based computation yields a single shortest distance, whereas volume-based computation yields a volume (blue region) that was not ablated with a prespecified margin (eg, 5 mm). In the left panel, no orange circle is visible because the tumor contour and expanded tumor contour coincide.

Volume-based Method

In the volume-based method, the tumor contour is first expanded by the desired MAM, and then the volume of the expanded tumor contour that lies outside the ablation volume contour is calculated (Fig 5, bottom) (41). This process can be repeated for various desired MAMs, such as 0, 5, and 10 mm. If the uncovered volume is 0 mL, this indicates that the desired MAM has been achieved. Similar to the distance-based method, the uncovered volume can be computed in discrete space using segmentation masks or in continuous space using 3D models. In the volume-based MAM (also called volume of insufficient coverage), a smaller uncovered volume corresponds to a larger margin.

Special Cases: Subcapsular and Perivascular Tumors

Ablation confirmation is notably challenging for subcapsular tumors. When the distance between the tumor and the liver capsule is smaller than the desired MAM, the computation of the minimal distance between the tumor and the liver capsule incorrectly identifies the MAM as the intraparenchymal ablation volume. By definition, the ablation volume should not extend beyond the liver capsule. Two methods have been proposed to address this issue (Fig 6, top row). The first method manually extends the ablation volume contour outside the liver by 10 mm (41) (Fig 6, top row, middle). In the second method, portions of the tumor surface that are closer to the liver capsule than the desired MAM are excluded from the MAM computation (43) (Fig 6, top row, right). Real examples of this scenario are included in Figure S1. A similar issue arises with perivascular tumors, in which the ablation volume cannot overlap major vessels, resulting in insufficient margins. The same adjustments used for MAM computation in subcapsular tumors can also be applied to perivascular tumors (Fig 6, bottom row).

Figure 6:

Illustration of the measurement of the minimal ablative margin (MAM) in subcapsular and perivascular tumors (red). (Top row, left) The ablation volume (blue) cannot grow beyond the capsule, so in the case of a subcapsular tumor, measurement of the MAM might be inaccurate. (Middle) An alternative approach that artificially expands the ablation volume outside the liver (41). (Right) Another alternative approach that removes tumor voxels from the computation (43). (Bottom row, left) In perivascular tumors, the ablation volume cannot overlap major vessels and typically extends up to the vascular wall, which may lead to an inaccurate MAM. (Middle) An alternative approach in which the vasculature is included in the ablation volume. (Right) Another alternative approach in which the tumor voxels close to the vasculature are removed from MAM computation. With both approaches, visual confirmation that the ablation extended all the way to the vessel is still required.

Sources of MAM Measurement Error

To achieve a unified frame of reference for measuring the MAM, several image processing steps are necessary, each introducing inaccuracies into the MAM measurement. These inaccuracies can accumulate or cancel each other out, and in certain circumstances, they can have a multiplicative effect. For example, inaccuracies in the segmentation process can amplify errors in a contour-based registration method. These various sources of error must be considered when interpreting MAM results. Unfortunately, there is currently very little evidence of the effects of these sources of error on the definition of an A0 ablation.

Image Resolution

In CT-guided procedures, image resolution typically falls to approximately 1 mm in the axial plane, whereas section thickness varies between 1 and 5 mm (18,19,44). Consequently, the precision of margin measurements is not uniform; accuracy is lower in the craniocaudal direction compared with the mediolateral and anteroposterior directions (Fig S2). However, image resolution impacts various aspects, including the accuracy of segmentation and registration. Thinner sections may exhibit a lower signal-to-noise ratio, affecting both human radiologists’ and artificial intelligence (AI) models’ ability to accurately segment the tumor and ablation zone using two-dimensional section-by-section methods as well as 3D methods (45). Low signal-to-noise ratios can also adversely affect intensity-based registration algorithms. In the absence of clinical evidence, section thickness should generally be smaller than the desired margin but large enough to achieve a sufficient signal-to-noise ratio for accurate image processing.

Segmentation Accuracy and Interreader Variability

Although automatic segmentation algorithms have drastically improved over the last decade, careful review and manual postprocessing by a radiologist are still necessary. However, this process introduces operator bias, leading to inter- and intrareader variability. Well-contrasted images and high signal-to-noise ratios (eg, thicker sections) can mitigate operator bias and enhance the baseline accuracy of automatic segmentation. Additionally, most segmentation algorithms tend to detect false-positive findings, especially in patients who have undergone previous ablation treatments, because previously ablated areas often appear as hypodense regions. However, such false positives can be easily excluded through imaging review.

Registration Accuracy

The accuracy of the registration process is likely the most critical factor influencing the overall accuracy of MAM quantification, as shown in Figure 7. Therefore, it is crucial for the operator to verify that the registration is accurate. In contour-driven registration methods, the segmentation of the liver directly affects the registration. Consequently, segmentation inaccuracies can lead to registration inaccuracies. The most common landmarks for verifying registration include vessel bifurcations, the liver surface, and, if available, fiducials, surgical clips, or liver parenchymal calcifications. Accuracy can then be assessed by using multicolor overlays or directly measuring the distances between these landmarks.

Figure 7:

CT during hepatic arteriography image shows the effect of inaccurate registration. Significant deformation between the pre- and postablation with intensity-based rigid registration: the tumor is inaccurately registered, and the minimal ablative margin (MAM) is misjudged. The rigid registration shows insufficient margins, whereas the deformable registration shows complete coverage with a 5-mm MAM. The rigid mapping of the liver (white) shows that the rigid registration was unable to correct the deformation in the left lobe, which was introduced by the ablation probe insertion and the ablation itself.

Deformation

The liver’s shape can change due to various factors (46), posing substantial challenges for registration. Patient positioning and breathing motion are also pertinent issues in stereotactic thermal ablation, which can be addressed through patient immobilization, the use of intraprocedural images acquired under general anesthesia, and breath holds or high-frequency jet ventilation (47). Implementing these techniques in conventional CT-guided ablation procedures can reduce deformations and improve the accuracy of MAM measurements, especially when deformable registration is not available.

To protect nearby critical structures, such as the diaphragm and bowel, maneuvers like hydrodissection can create a space between the liver and these structures. This prevents heat damage but can also lead to significant liver deformation, negatively affecting MAM measurements. If hydrodissection is employed, the preablation CECT scan should ideally be acquired after the hydrodissection, particularly if no biomechanical model is used.

During the thermal ablation process, liver tissue contracts due to dehydration and tissue shrinkage (48). It is generally assumed that as tissue shrinks, the tumor itself also reduces in size. However, there is currently no evidence regarding the relationship between intra- and interpatient normal liver shrinkage and tumor shrinkage. Therefore, a conservative estimate of no tumor tissue shrinkage is warranted until better data are available. In a study by Vasiniotis Kamarinos et al (20), a tissue contraction algorithm developed by one of the microwave ablation device manufacturers showed significantly worse sensitivity in detecting local tumor progression compared with using the same software without the tissue contraction algorithm. This is likely because the algorithm assumes isometric tissue shrinkage toward the center of the tumor, regardless of the ablation probe’s location. Additionally, tissue shrinkage at the liver capsule results in substantial changes in liver shape (Fig 8). A potential solution is to use deformable registration methods that estimate the deformation field across the liver, because rigid registration often leads to the tumor being mapped outside the liver (49).

Figure 8:

CT images during hepatic arteriography in a 55-year-old woman with intrahepatic cholangiocarcinoma who underwent percutaneous thermal ablation for four tumors. (A) One tumor measuring 1.8 cm was located close to the liver capsule in segment VI and was ablated using one probe with 65 W for 10 minutes (arrowhead). (B) The intraprocedural postablation image shows significant deformation at the liver capsule (arrowhead) caused by the ablation process.

Image Artifacts

Many patients undergoing thermal ablation have previously undergone ablation or surgery; thus, surgical clips, fiducial markers, and other implants that can produce image artifacts are common. In patients with intrahepatic cholangiocarcinoma, biliary drains and stents can also cause image artifacts. The ablation probe itself introduces beam-hardening artifacts and should, therefore, be at least partially retracted outside the ablation zone for postablation imaging. Additionally, components of stereotactic or robotic systems, such as aiming devices, should also be fully moved outside the field of view to prevent metal artifacts (Fig S3).

Overview of Ablation Confirmation Software Solutions

Here, we provide an overview of available ablation confirmation software solutions, some of which are commercially available, in terms of modes of integration. It is important to note that there is no evidence of improved outcomes associated with one solution over another. Additionally, the landscape of commercial solutions evolves rapidly as new options are developed and updated. However, recent consensus guidelines suggest that the use of ablation confirmation software is highly desirable (15) and performs better than visual inspection and manual measurements, which have been shown to be unreliable (14). A randomized clinical trial demonstrated that the intraprocedural use of an ablation confirmation software significantly improved thermal ablation efficacy by increasing the proportion of patients achieving the optimal MAM, as well as enhancing local tumor control, when compared to visual inspection (66).

Levels of Integration Used in Software Solutions

Stand-Alone Software

Most stand-alone software solutions employ AI-based auto-segmentation techniques in addition to manual tools for segmenting the liver, tumors, and ablation zones. Some software solutions also provide auto-segmentation of the vasculature. The main differences among stand-alone software solutions lie in the registration techniques applied and additional functionalities offered, such as planning and probe position confirmation. The most common registration techniques include intensity-based deformable (Ablation-fit [R.A.W.] [21,44,50], IQQA BodyImaging [EDDA Technology], MIM MAESTRO [MIM Software]), intensity-based rigid (BioTrace IO [Techsomed]) (51), and biomechanical deformable (RayStation [Raysearch Laboratories]) (19) registration. Figure 9 shows an example of stand-alone software using AI-based auto-segmentation and biomechanical deformable image registration based on a radiation treatment planning system, which has been used in the recently published COVER-ALL trial (66), as well as another ongoing prospective study (52).

Figure 9:

CT images in a 52-year-old man with colorectal liver metastases who underwent percutaneous thermal ablation of two tumors (green) that were both deemed completely ablated with sufficient margins per conventional image fusion. (A) For the segment VIII tumor measuring 1.7 cm, (B) deformable registration of the postablation contrast-enhanced CT (CECT) image shows insufficient margins (minimal ablative margin [MAM] = 0, red) on the medial anterior and posterior aspect of the ablation zone (orange). (C) Follow-up CECT image 1 year after the procedure shows local tumor progression at the location of the insufficient margin (arrowhead). (D) For the tumor in segment IV measuring 1.2 cm, (E) deformable registration of the postablation CECT image shows complete ablation with a greater than 5 mm MAM surrounding the tumor. (F) Follow-up CECT image 1 year after the procedure shows no local tumor progression.

Integrated Software

Manufacturers of imaging and image-guidance systems, as well as ablation devices, have integrated ablation confirmation functionalities into their software workflows. Consequently, most centers will soon have ablation confirmation functionalities readily available on one of these systems. Some centers will also offer this functionality across multiple systems (eg, on a stereotactic or robotic device and the ablation device). Therefore, interoperability among these systems will become increasingly important, highlighting the need for standardized protocols that include data such as trajectories and ablation power settings.

Interventional imaging systems, particularly CT, have integrated procedure planning, image fusion, and MAM assessment functionality directly into the CT workstations. This includes semiautomatic segmentation and registration methods, such as myAblation Guide (Siemens Healthineers) (Fig 10). Because these workstations are part of the essential imaging equipment, no external software or hardware is required, simplifying setup and data transfers.

Figure 10:

CT images in a 64-year-old man with an American College of Radiology Liver Imaging Reporting and Data System 4 lesion suspicious for hepatocellular carcinoma who underwent percutaneous thermal ablation. (A) A 1.9-cm lesion in segment V delineated (green) with the intended 5-mm minimal ablative margin (yellow) on the preablation contrast-enhanced CT (CECT) image. (B) Image fusion with the preablation CECT demonstrates accurate placement of the ablation probe. (C) Postablation CECT intensity-based rigid registration-based image shows accurate fusion on the colored overlay. (D, E) After ablation zone segmentation (orange), the tumor (green) is shown to be completely covered but the margin (yellow) on the inferior aspect is shown to be insufficient (orange and red, arrowhead). (F) Follow-up CECT image 8 months after the procedure shows no local tumor progression.

Stereotactic and robotic image-guidance systems, like CAS-One (CAScination) and Epione (Quantum Surgical), employ patient-to-image and image-to-image registration, along with organ segmentation for the stereotactic workflow. Newer versions of these devices have added functionalities for pre- and postablation CECT image fusion for margin quantification. Given that stereotactic and robotic guidance relies on rigidity, images are acquired under controlled breathing conditions, and patients are immobilized using vacuum mattresses or other devices. Consequently, these systems currently rely on rigid registration techniques, which do not address all sources of MAM quantification errors. However, with the use of high-frequency jet ventilation during stereotactic guidance, organ motion is minimized. Therefore, simple and fast rigid registration techniques can still yield reliable MAM estimates, even in cases with minimal deformation due to the ablation process itself (Fig 11).

Figure 11:

CT images in a 73-year-old woman with hepatocellular carcinoma who underwent stereotactic percutaneous thermal ablation under general anesthesia using high-frequency jet ventilation. (A) A single trajectory was planned to ablate one segment VIII tumor measuring 1.7 cm (red) with a 5-mm minimal ablative margin (MAM) (orange). (B) The ablation probe position was confirmed to be placed as planned. (C) Postablation contrast-enhanced CT (CECT) rigid registration-based fusion using image coordinates shows minimal organ deformation that does not require any manual adjustments. (D, E) The tumor and 5-mm MAM were quantitatively confirmed to be covered 100% by the ablation zone (green) and visualized on axial planes as well as three-dimensional reconstruction. (F) Follow-up CECT image 2 months after the procedure showed no local tumor progression.

The Neuwave ablation system (NeuWave Medical) features built-in ablation confirmation software with a graphical user interface displayed on the device screen (Fig 12). This software employs both automatic and semiautomatic segmentation techniques for the tumor and ablation volume. For registration, it used intensity-based deformable registration to align pre- and postablation CT scans. Additionally, the software incorporates a proprietary algorithm to account for tissue contraction, a common issue arising from microwave ablation. As previously described, a recent study found a decrease in sensitivity for detecting suboptimal margins when this algorithm was activated (20).

Figure 12:

CT images in a 72-year-old woman with colorectal liver metastases undergoing percutaneous thermal ablation of one tumor. (A) Intraprocedural preablation contrast-enhanced CT (CECT) image shows a 1.2-cm tumor in segment VII was loaded onto the ablation planning and confirmation software integrated on the ablation device with the tumor (red) and the intended 5-mm minimal ablative margin (MAM) segmented (pink). (B) Accurate ablation probe positioning was confirmed, and the estimated ablation zone (orange) predicts complete coverage of the tumor and 5-mm MAM. (C) Postablation CECT rigid registration-based fusion image shows complete coverage of the 5-mm MAM (pink). (D) Follow-up CECT image 1 month after the procedure shows no local tumor progression.

Evidence from Clinical Studies

Efficacy of Thermal Ablation with Sufficient Margins

Although studies reporting overall local tumor progression (LTP) rates after thermal ablation show wide variability, studies reporting LTP rates based on MAM show more consistent results. Recently, our group published the COVER-ALL Trial (NCT04083378), the first randomized, phase 2, superiority study evaluating a novel software-based method for MAM assessment incorporating biomechanical deformable image registration and AI-based autosegmentation during CT-guided thermal ablation of liver tumors (66). Patients were randomized intraprocedurally (1:1) to either software-based or conventional visual MAM assessment. All tumors were treated with the intent of achieving an MAM of at least 5 mm. The primary end point was MAM at postablation CT. A preplanned interim analysis after 50% enrollment demonstrated a significantly larger mean MAM in the experimental group (5.9 mm [SD 2.7]) compared to controls (2.2 mm [2.8]; P < .0001), leading to early cessation of randomization. The intended MAM of 5 mm or larger was achieved in 75% of the tumors treated in the randomized experimental group and only in 15% of the tumors treated in the control group. Subsequently, 50 additional patients were enrolled into a nonrandomized experimental group, achieving a mean MAM of 7.2 mm (±2.8) with a MAM of at least 5 mm achieved in 84% of the treated tumors. Across 100 patients, only grade 1–3 adverse events occurred in 5% of cases, with no grade 4–5 events or treatment-related deaths. The findings of this trial indicate that intraprocedural software-based MAM assessment is safe and significantly improves ablation quality compared to visual assessment, supporting its integration into standard thermal ablation workflows (66). Another small study prospectively assessed MAM in individuals undergoing thermal ablation for hepatocellular carcinoma; this study reported a 0% LTP rate with an MAM of 0 mm or more (23). However, the primary end point of this study was the technical feasibility of the software used. Retrospective studies reported LTP rates for tumors ablated with a MAM greater than 5 mm, ranging from 0% to 8.4% for hepatocellular carcinoma and from 0% to 12% for colorectal liver metastases. These LTP rates are summarized in Table 2.

Table 2:

Summary of Recent Studies Involving Use of Quantitative 3D Software for MAM Assessment after Percutaneous Thermal Ablation of Liver Tumors

What Is the Optimum Margin?

To date, there is no definitive consensus on the MAM that should be interpreted as indicating an A0 ablation. The two main factors influencing this definition are the tumor histologic subtype and the measurement technique. Most of the available evidence pertains to the ablation of colorectal liver metastases and hepatocellular carcinoma, with sufficient MAM reported to range from 2 to 10 mm. The lower thresholds primarily arise from studies involving highly accurate 3D measurement methods and very strict inclusion criteria (eg, no tissue deformation) (18,44), whereas the higher thresholds stem from studies using manual measurements on two-dimensional planes with anatomic landmarks (13,54,55). Thus, the more accurate measurement methods must account for less measurement error in establishing an A0 definition and may potentially allow for smaller margins. A recent meta-analysis that attempted to compute the LTP rate stratified by margins incorporated all measurement techniques (visual assessment and software-based) and found significant heterogeneity among the studies (56), which limits the identification of optimum margins. Table 2 presents an overview of recent studies involving dedicated quantitative margin assessment software by employing tumor and ablation segmentation, image fusion, and quantitative measurements.

Research into novel technologies has primarily focused on increasing the MAM to improve A0 rates. As these rates rise, the definition of MAM will likely evolve into a range of margins to avoid unnecessary destruction of healthy liver parenchyma, which could be linked to higher complication rates and hinder future salvage ablation treatments. More tumor-specific trials, such as the ACCLAIM ([Ablation with Confirmation of Colorectal Liver Metastases] ClinicalTrials.gov: NCT05265169) and PROMETHEUS (Prostate Multicenter External Beam Radiotherapy Using Stereotactic Boost) (57) trials, are needed to establish optimum histology-specific MAM cutoffs and ranges.

Management of Insufficient Margins

Similar to R2 resections, A2 ablations (MAM ≤ 0 mm, ie, tumor left behind) should be retreated within the same procedure whenever possible, because LTP rates for MAM ≤ 0 mm can be as high as 100% (58). For insufficient margins (A1, MAM between 0 and 5 mm) the risk of immediate reablation must be carefully weighed against the potential benefit of increasing MAM—especially for insufficient MAMs close to critical structures. A recent multicenter retrospective study indicated that there are diminishing returns for local tumor control once the MAM reaches 4.2 mm, whereas there are significant potential gains with MAMs in the range of 0–2.5 mm (59). Therefore, efforts should be made to achieve A0 intraprocedurally on the first attempt, and the use of ablation confirmation software during the procedure is strongly recommended. In the COVER-ALL trial, a higher frequency of reablation and overlapping ablations were performed on the experimental (software-based MAM assessment) randomized group compared to the control (visual inspection) group (42% vs 19%, P = .084 and 92% vs 62%, P = .013, respectively). This indicates the relevance of software-based ablation margins confirmation methods as an actionable intraprocedural decision-making tool. The addition of other local-regional techniques (eg, embolization or external radiation) to enhance insufficient margins should be avoided outside clinical trials until clear evidence becomes available, because they might limit future curative-intent salvage local-regional therapies. A shorter imaging follow-up to detect and reablate residual unablated tumor or LTP early may be a more appropriate strategy.

Impact of Ablation Confirmation Software–based MAM Quantification on Other Predictors of Local Outcomes

A review of the evidence regarding MAM and other risk factors for recurrence after thermal ablation, particularly ablation of colorectal liver metastases, indicates that risk factors identified without, or with less accurate, MAM measurement were not significant in newer studies using more accurate MAM measurement techniques. For instance, RAS mutations were previously identified as being associated with higher LTP rates after thermal ablation of colorectal liver metastases (54,60,61), with studies further suggesting that the MAM for RAS-mutated tumors should be higher than that for RAS–wild-type tumors (55). However, in more recent studies employing more accurate MAM measurements with AI-based segmentation and biomechanical deformable registration instead of manual measurements using anatomic landmarks, RAS mutation was no longer identified as a significant risk for LTP (18,19). Although stronger evidence is still warranted, these recent findings underscore the importance of margin assessment using accurate quantitative 3D software. A recently published randomized controlled trial tested whether ablation confirmation software improves margins and demonstrated that the intraprocedural use of this software significantly increased the proportion of tumors achieving adequate MAMs (66).

Future Outlook and Conclusion

To date, one of the major barriers to the widespread use of ablation confirmation software during thermal ablation of malignant liver tumors is the lack of reimbursement, which limits the software’s use to large academic centers. Furthermore, as previously described, there is persistent heterogeneity among systems in terms of technical design, type of registration, and varying levels of operator supervision. This raises questions concerning the reproducibility of the results and highlights the need for standardization in this emerging field.

Additionally, even with state-of-the-art AI-enabled software, image processing takes 5 to 10 minutes due to the need for image transfers to dedicated software, manual verification and correction steps, and computation time. To preserve the option of immediate reablation, patients should remain in the interventional radiology suite, under general anesthesia, until ablative margins are confirmed. Furthermore, more technical developments are necessary to reduce computation time and improve interoperability between systems. Thus, in addition to software license costs, the additional requirements for intervention room and professional time increase the overall cost of the procedure. Alternative or separate reimbursement models would enable smaller, especially private centers, to cover these costs. Large multicenter, multinational trials, such as ACCLAIM (NCT05265169), are ongoing and may provide the evidence necessary for insurers to cover these additional expenses. Moreover, the majority of current evidence is limited to hepatocellular carcinoma and colorectal liver metastases, but ablation is increasingly being used for other types of metastases. Therefore, it is crucial to identify and validate criteria for an A0 ablation for other histologic subtypes.

In addition, management strategies for insufficient margins should be studied further. Currently, the evidence suggests that insufficient MAMs should be corrected intraprocedurally with additional ablation whenever possible. In this context, proper patient selection and the use of ablation confirmation methods are of critical importance. The role of adjuvant systemic treatment for addressing potential residual disease in cases of insufficient margins also needs further investigation. Moreover, the use of MAM for modifying follow-up schedules, in conjunction with other biomarkers (eg, circulating tumor DNA), should be explored further.

In conclusion, data regarding ablation confirmation software and MAM quantification have shown very promising results based on retrospective studies and a recently published randomized trail. Ongoing histology-specific prospective studies will further strengthen the ongoing evidence for ablation confirmation with intraprocedural MAM assessment as the new standard of care for liver ablation in various types of primary and secondary liver cancer, thereby enhancing the relevance and reproducibility of thermal ablation as an effective, curative-intent local therapy.