Identification of A0 minimum ablative margins for colorectal liver metastases: multicentre, retrospective study using deformable CT registration and artificial intelligence-based autosegmentation

Abstract

Background

Several ablation confirmation software methods for minimum ablative margin assessment have recently been developed to improve local outcomes for patients undergoing thermal ablation of colorectal liver metastases. Previous assessments were limited to single institutions mostly at the place of development. The aim of this study was to validate the previously identified 5 mm minimum ablative margin (A0) using autosegmentation and biomechanical deformable image registration in a multi-institutional setting.

Methods

This was a multicentre, retrospective study including patients with colorectal liver metastases undergoing CT- or ultrasound-guided microwave or radiofrequency ablation during 2009–2022, reporting 3-year local disease progression (residual unablated tumour or local tumour progression) rates by minimum ablative margin across all institutions and identifying an intraprocedural contrast-enhanced CT-based minimum ablative margin associated with a 3-year local disease progression rate of less than 1%.

Results

A total of 400 ablated colorectal liver metastases (median diameter of 1.5 cm) in 243 patients (145 men; median age of 62 [interquartile range 54–70] years) were evaluated, with a median follow-up of 26 (interquartile range 17–40) months. A total of 119 (48.9%) patients with 186 (46.5%) colorectal liver metastases were from international institutions B, C, and D that were not involved in the software development. Three-year local disease progression rates for 0 mm, >0 and <5 mm, and 5 mm or larger minimum ablative margins were 79%, 15%, and 0% respectively for institution A (where the software was developed) and 34%, 19%, and 2% respectively for institutions B, C, and D combined. Local disease progression risk decreased to less than 1% with an intraprocedurally confirmed minimum ablative margin greater than 4.6 mm.

Conclusion

A minimum ablative margin of 5 mm or larger demonstrates optimal local oncological outcomes. It is proposed that an intraprocedural minimum ablative margin of 5 mm or larger, confirmed using biomechanical deformable image registration, serves as the A0 for colorectal liver metastasis thermal ablation.

The aim of this study was to validate the previously identified 5 mm minimum ablative margin (A0) after thermal ablation of colorectal liver metastases using biomechanical deformable image registration and autosegmentation in a multi-institutional setting. This retrospective study included patients with colorectal liver metastases undergoing CT- or ultrasound-guided microwave or radiofrequency ablation during 2009–2022 from four institutions (the institution where the software was developed and three international institutions that were not involved with software development). Three-year local disease progression rates for 0 mm, <0 and <5 mm, and 5 mm or larger minimum ablative margins were 79%, 15%, and 0% respectively for the institution where the software was developed and 34%, 19%, and 2% respectively for the three other institutions combined.

Introduction

Percutaneous thermal ablation, a minimally invasive curative-intent locoregional therapy for patients with colorectal liver metastases (CRLMs), can be used alone or in combination with surgery^1,2. Historically, percutaneous thermal ablation has been used predominantly for patients with small (less than 3 cm) and limited (less than 3) CRLMs not amenable to surgery. However, thermal ablation for smaller CRLMs is increasingly being used as first-line treatment because of the lower cost, fewer complications, and faster recovery³ and might further increase given the recent findings of a randomized trial comparing first-line resection versus ablation^4,5.

Complete tumour coverage with a sufficient minimum ablative margin after thermal ablation has been shown to be the most important predictor of local tumour progression^6–11. In the past, minimum ablative margin was measured using anatomical landmarks¹² by image co-registration with visual inspection, which was shown to be inaccurate even when performed by experienced physicians¹³. Older studies relying on anatomical landmarks reported an optimal minimum ablative margin of more than 10 mm^12,14, whereas newer studies using specialized three-dimensional (3D) software reported an optimal minimum ablative margin of 2 to 5 mm or larger^6,7,10,11. In addition, other factors previously linked with local tumour progression, such as carcinoembryonic antigen, RAS mutation status, and CRLM location, were no longer statistically significant predictors of local tumour progression when a 3D software-determined minimum ablative margin was included in the analysis^10,15, emphasizing the importance of using 3D software analysis to confirm minimum ablative margin.

A minimum ablative margin quantification method using biomechanical deformable image registration (DIR) and artificial intelligence (AI)-based autosegmentation has been previously developed^16,17. This method has been validated in a large retrospective cohort⁶ and in an ongoing phase 2 randomized trial¹⁸. In the retrospective cohort, it was found that minimum ablative margin was the most important predictor of local tumour progression and that no CRLMs with a minimum ablative margin of 5 mm or larger had local tumour progression. Like most other recently developed ablation confirmation methods, it has only been validated at its institution of development or in small samples, which limits the generalizability of these findings^6,10,11,19.

The aim of this multi-institution study was to validate an ablation confirmation software for minimum ablative margin quantification equipped with AI-based autosegmentation and biomechanical DIR on data sets with patient populations and ablation techniques different from those at the institution where the software was developed. It was hypothesized that to achieve a local disease progression (residual tumour or local tumour progression) rate of less than 5% (A0) and an upper 95% confidence interval of less than 10%, ablated CRLMs from institutions B, C, and D (see below) would require a minimum ablative margin of 5 mm or larger. Another aim was to estimate a precise minimum ablative margin threshold that minimized risk of local disease progression to 5% or less and beyond which each 1-mm increase in minimum ablative margin resulted in a less than 1% decrease in local disease progression risk.

Methods

Study design

This multi-institution retrospective analysis included four institutions: the institution where the software was developed (institution A) and three other institutions from three other countries (institutions B, C, and D). This study was approved by the respective institutional review boards and received a waiver for informed consent (A, 2021-0340; B, 2021-0428; C, AN4357; and D, 674.586/00001.2022). Overlaps between this study’s patient population and previous publications are disclosed in the Supplementary material.

Selection criteria

Consecutive patients submitted for curative-intent microwave or radiofrequency percutaneous thermal ablation of CRLMs between January 2009 and December 2022 were included. Patients with up to five CRLMs measuring 5 cm or less were included. Patients with inaccurate image registration, any other locoregional therapy (for example resection or chemoembolization) to the target lesion(s) concurrent with or before ablation, less than 1 year of imaging follow-up and no local disease progression, no contrast-enhanced CT (CECT) within 60 days of the ablation, or insufficient CECT image quality for tumour or ablation zone depiction were excluded (Fig. 1). This study was conducted in accordance with STROBE guidelines²⁰.

Fig. 1.

Study flow chart summarized across all institutions

*Residual tumour was detected at the end of the procedure, but re-ablation was deferred to a later ablation session. LRT, locoregional therapy, LTP, local tumour progression, CRLMs, colorectal liver metastases.

Data collection

Clinical and oncological follow-up data from all institutions were deidentified and entered into the study database managed centrally by researchers at institution A via material transfer agreements. Baseline characteristics, including the presence of extrahepatic disease, lung metastases, and positive lymph nodes, were collected from the electronic medical records. For each CRLM, its diameter, liver segment, and perivascular and subcapsular location were collected from the picture archiving and communication system.

Outcome assessment

All oncological outcomes were assessed at each institution by local investigators (trained radiologists with 5–26 years of experience) who were blinded to the results of the minimum ablative margin quantification. Local oncological outcomes for institutions B, C, and D were validated at institution A and disagreements on outcomes were resolved by consensus between investigators from institution A and the respective institutions.

Ablation outcomes were recorded according to standardized terminology based on post-ablation cross-sectional imaging (CECT, MRI, or PET)^21,22. Residual tumour was defined as tumour foci within or at the edge of the ablation zone at initial follow-up imaging. Local tumour progression was defined as tumour foci within or at the edge of the ablation zone after at least one cross-sectional imaging study had demonstrated complete ablation. Local disease progression was defined as the presence of residual tumour or local tumour progression. Time to local disease progression was defined as the time between ablation and the first imaging evidence of residual tumour or local tumour progression. Time to development of new CRLMs was defined as the time between the ablation procedure and the appearance of new tumour foci within the liver, but not in contact with the ablation zone. Overall survival (OS) and time to development of new CRLMs were measured from the first ablation session.

Minimum ablative margin quantification

For quantification of minimum ablative margin, the authors used their developed ablation confirmation methodology, which encompasses radiation oncology planning software (Raystation, Raysearch Laboratories, Stockholm, Sweden) equipped with a biomechanical DIR coupled with a custom-built AI segmentation algorithm. This ablation confirmation methodology for minimum ablative margin quantification was developed and extensively evaluated previously at institution A⁶ and is currently being used in two prospective trials at institution A^18,23.

Deidentified imaging data were transferred via an institutional secure file transfer server and then imported into the minimum ablative margin assessment software. All images were initially processed using the automated segmentation by a trained biomedical engineer and then verified by a trained radiologist. Then, the pre- and post-ablation images were registered using the biomechanical DIR and the registration accuracy was verified by the trained radiologists. In cases of inaccurate registration, the registration was repeated alongside an engineer to rule out technical issues. To avoid operator bias, no manual corrections to registration were allowed. When the trained radiologist deemed the registration inaccurate by visual inspection using overlays, after repeated (maximum of 5 attempts) registration attempts, the procedure was excluded. The minimum ablative margin was then computed as the shortest distance between the ablation and the tumour contour and categorized as 0 mm, >0 and <5 mm, and 5 mm or larger. As less than 5% of CRLMs had a minimum ablative margin of 10 mm or larger, a separate category for margins more than 10 mm was not included.

Differences in ablation procedures and image acquisitions

Institutions A, B, C, and D used different image-guidance techniques for liver tumour ablation (see the Supplementary material). All procedures were performed with curative intent and with patients under general anaesthesia (institutions A, B, and C) or deep sedation (institution D) with the goal of completely covering the tumours with a margin of 5 mm or more. Complete coverage of the tumour was assessed visually without the aid of dedicated, quantitative margin assessment software.

A0 margins and A0 ablation

An A0 margin was defined as a minimum ablative margin with a 3-year local disease progression risk of less than 5%. Additionally an A0 ablation was defined as an ablation with an A0 margin. Further, the point of diminishing returns was defined as the minimum ablative margin threshold where additional 1-mm increases in the minimum ablative margin yielded a less than 1% decrease in risk for local disease progression.

To determine the threshold for A0 margins, first, a logistic regression model for detecting 3-year local disease progression according to minimum ablative margin was fitted. Then, the 3-year local disease progression risk for each minimum ablative margin threshold between 0 and 10 mm was estimated to identify the precise threshold where local disease progression risk dropped below 5%. In addition, the 95% confidence interval of the prediction was computed to report the uncertainty of the minimum ablative margin threshold. The derivative of this local disease progression risk function was computed to find the point where a 1-mm increase in minimum ablative margin yielded a less than 1% decrease in local disease progression risk. This analysis was performed for cases in which both pre- and post-ablation CECT images were acquired intraprocedurally and cases in which either or both were acquired periprocedurally.

Statistical analysis

Statistical analysis was performed using R (version 4.2.2) in RStudio (version 2022.12.0)²⁴, with the packages ggplot2 and ggpubr for graphics^25,26, and tidyverse was used for data processing²⁷. The R code is available on Zenodo (https://zenodo.org/doi/10.5281/zenodo.10606186)²⁸.

Baseline patient and tumour characteristics are reported using descriptive statistics as median (interquartile range) or n (%). Cumulative incidence curves were used to estimate rates of local disease progression and new CRLMs development at 3 years across institutions. OS was estimated using the Kaplan–Meier method. The log rank test was used for comparison. The area under the receiver operating characteristic (AUROC) curve was used to estimate the predictive performance of the minimum ablative margin for 3-year local disease progression and the DeLong et al.²⁹ method was used to calculate 95% confidence intervals. This analysis was performed by institution and imaging timing (intraprocedural or periprocedural). Patients lost to follow-up who did not experience local disease progression within 3 years were excluded from this AUROC calculation. A sensitivity analysis was performed for local disease progression within 1 and 2 years after ablation. All analyses were performed on a per-tumour level. A sensitivity analysis was performed using a mixed-effects Cox proportional hazards model with and without the patient and institution as a random effect. Additionally, per-patient cumulative incidence curves for local disease progression are reported in the Supplementary material. P < 0.050 was considered statistically significant and two-sided 95% confidence intervals are reported.

Results

Patient characteristics

A total of 243 patients (145 men; median age of 62 (interquartile range 54–70) years) with 400 ablated CRLMs in 287 ablation sessions were included in this study (Fig. 1). Of the 287 ablation sessions, 226 (78.7%) were performed in ablation-naive patients. Detailed patient and procedure characteristics are shown in Table 1 and Table 2 respectively. Patients from institutions B–D were on average 5 to 10 years older and had significantly less extrahepatic disease than patients from institution A. Patients from institutions C and D had 6 to 8 mm larger tumours on average than patients from institutions A and B.

Table 1.

Patient baseline characteristics across all institutions

Characteristic	Overall (n = 243)	Institution A (n = 124)	Institution B (n = 20)	Institution C (n = 52)	Institution D (n = 47)
Age (years), median (interquartile range)	62 (54–70)	57 (4965)	62 (59–73)	67 (60–73)	65 (59–77)
Sex
Male	145 (60)	69 (56)	10 (50)	37 (71)	29 (62)
Female	98 (40)	55 (44)	10 (50)	15 (29)	18 (38)
Extrahepatic disease	131 (54)	92 (74)	6 (30)	14 (27)	19 (40)
Lung metastases	68 (28)	45 (36)	2 (10)	6 (12)	15 (32)
Positive lymph nodes	90 (37)	73 (59)	3 (15)	10 (19)	4 (8.5)
Follow-up time (months), median (interquartile range)	26 (17–40)	25 (16–37)	42 (24–55)	33 (23–50)	25 (15–34)

Values are n (%) unless otherwise indicated.

Table 2.

Procedure characteristics across all institutions

Characteristic	Overall (n = 287)	Institution A (n = 152)	Institution B (n = 22)	Institution C (n = 61)	Institution D (n = 52)
Maximum tumour diameter (mm), median (interquartile range)	15 (12–23)	15 (10–20)	15 (14–25)	23 (15–33)	19 (15–24)
Number of ablated CRLMs, median (interquartile range)	1 (1–2)	1 (1–2)	1 (1–2)	1 (1–2)	1 (1–2)
Image guidance modality
CT	241 (84)	152 (100)	22 (100)	61 (100)	6 (12)
Ultrasonography	46 (16)	0 (0)	0 (0)	0 (0)	46 (88)
Ablation modality
MWA	176 (61)	144 (95)	0 (0)	0 (0)	32 (62)
RFA	111 (39)	8 (5.3)	22 (100)	61 (100)	20 (38)
Intraprocedural pre-ablation CECT	125 (44)	64 (42)	0 (0)	61 (100)	0 (0)
Intraprocedural post-ablation CECT	212 (74)	123 (81)	21 (95)	61 (100)	7 (13)

Values are n (%) unless otherwise indicated. CRLMs, colorectal liver metastases; MWA, microwave ablation; RFA, radiofrequency ablation; CECT, contrast-enhanced CT.

Oncological outcomes

The rates of development of new CRLMs within 3 years were 46% (95% c.i. 16% to 65%), 36% (95% c.i. 19% to 49%), and 48% (95% c.i. 30% to 62%) for institutions B, C, and D respectively and were significantly lower than the rate for institution A (63% (95% c.i. 51% to 73%); P < 0.001) (Fig. 2a and Table 3). OS was similar across all four institutions (P = 0.850), with median OS times of 48.8 (95% c.i. 40.9 to not reached), 52.3 (95% c.i. 41.1 to not reached), 66.5 (95% c.i. 46.0 to 89.5), and 49.8 (95% c.i. 37.1 to not reached) months for institutions A, B, C, and D respectively (Fig. 2b).

Fig. 2.

Intrahepatic new tumor incidence and overall survival Kaplan-Meier curves

a Cumulative incidence of development of new colorectal liver metastases after the initial ablation procedure for colorectal liver metastases. b Overall survival after the initial ablation procedure for colorectal liver metastases. CRLMs, colorectal liver metastases.

Table 3.

Tumour and ablation outcome characteristics across all institutions

Characteristic	Overall (n = 400)	Institution A (n = 214)	Institution B (n = 27)	Institution C (n = 90)	Institution D (n = 69)
Tumour diameter (mm), median (interquartile range)	15 (10–20)	12 (9–18)	15 (13–23)	20 (13–30)	18 (12–23)
Location
Subcapsular	167 (42)	98 (46)	14 (52)	30 (33)	25 (36)
Adjacent to vessel	120 (30)	55 (26)	12 (44)	29 (32)	24 (35)
MAM (mm), median (interquartile range)	3.00 (0.00–5.80)	4.30 (2.15–6.10)	2.20 (0.00–5.25)	2.05 (0.00–6.70)	0.00 (0.00–0.60)
MAM category (mm)
0	125 (31)	32 (15)	10 (37)	33 (37)	50 (72)
>0 and <5	145 (36)	94 (44)	10 (37)	25 (28)	16 (23)
5 or larger	130 (33)	88 (41)	7 (26)	32 (36)	3 (4.3)
Local disease progression	79 (20)	36 (17)	4 (15)	12 (13)	27 (39)
Residual tumour	21 (5.3)	7 (3.3)	2 (7.4)	4 (4.4)	8 (12)
Local tumour progression	58 (15)	29 (14)	2 (7.4)	8 (8.9)	19 (28)
Local disease progression within 3 years	79 (20)	36 (17)	4 (15)	12 (13)	27 (39)

Values are n (%) unless otherwise indicated. MAM, minimum ablative margin.

Three-year local disease progression

Local disease progression rates at 3 years for 0 mm, >0 and <5 mm, and 5 mm or larger minimum ablative margins were 79% (95% c.i. 56% to 90%), 15% (95% c.i. 7% to 23%), and 0% (95% c.i. 0% to 0%) respectively for institution A and 34% (95% c.i. 23% to 44%), 19% (95% c.i. 7% to 30%), and 2% (95% c.i. 0% to 7%) respectively for institutions B, C, and D combined (Fig. 3). Cumulative incidence curves according to minimum ablative margin category on a per-tumour level and per-patient level are provided in the Supplementary material for each institution. The HRs for local disease progression with >0 and <5 mm and 5 mm or larger minimum ablative margins were 0.29 (95% c.i. 0.18 to 0.48) and 0.03 (95% c.i. 0.01 to 0.1) in the Cox proportional hazards model without random effects and 0.23 (95% c.i. 0.13 to 0.41) and 0.02 (95% c.i. 0.00 to 0.08) with the patient and institution as a random effect (see the Supplementary material).

Fig. 3.

Cumulative incidence of local disease progression by minimum ablative margins

a Cumulative incidence of local disease progression according to minimum ablative margins for institution A. b Cumulative incidence of local disease progression according to minimum ablative margins for institutions B, C, and D. MAM, minimum ablative margin; LDP, local disease progression.

The AUROC values for predicting local disease progression within 3 years based on minimum ablative margin were 0.94, 0.51, 0.79, and 0.47 for institutions A, B, C, and D respectively. The AUROC value for predicting local disease progression within 3 years based on minimum ablative margin was higher for intraprocedural CECT (0.92 (95% c.i. 0.86 to 0.98)) than for periprocedural CECT (0.77 (95% c.i. 0.69 to 0.85)) (Fig. 4). Additional sensitivity analyses with receiver operating characteristic (ROC) curves by institution for predicting local disease progression based on minimum ablative margin within 1, 2, and 3 years are provided in the Supplementary material.

Fig. 4.

Receiver operating characteristic (ROC) curves for predicting 3-year local disease progression by minimum ablative margins measured using intraprocedural versus periprocedural contrast-enhanced CT imaging

CECT, contrast-enhanced CT; AUROC, area under the receiver operating curve.

The logistic regression model fitted to predict 3-year local disease progression had AUC values of 0.92 and 0.77 using intraprocedural imaging and periprocedural imaging respectively. Using intraprocedural imaging (185 CRLMs), the local disease progression risk was below 5% at a minimum ablative margin of 3.2 mm (95% c.i. 1.7 mm to 4.2 mm) and, once the minimum ablative margin reached 4.6 mm (95% c.i. 2.7 mm to 5.8 mm), the local disease progression risk decreased less than 1% per additional 1 mm of minimum ablative margin (Fig. 5a). Using periprocedural imaging (215 CRLMs), the local disease progression risk was below 5% at a minimum ablative margin of 7.0 mm (95% c.i. 4.4 mm to 9.0 mm) and, once the minimum ablative margin reached 8.9 mm (95% c.i. 6.4 mm to 11.0 mm), the local disease progression risk decreased less than 1% per additional 1 mm of minimum ablative margin (Fig. 5b).

Fig. 5.

Risk of local disease progression according to minimum ablative margins with thresholds for sufficient minimum ablative margins (less than 5% local disease progression risk)

a Measured using intraprocedural imaging. b Measured using periprocedural imaging. MAM, minimum ablative margin; DR, point of diminishing returns (the minimum ablative margin threshold where additional 1-mm increases in the minimum ablative margin yielded a less than 1% decrease in risk for local disease progression); LDP, local disease progression.

Discussion

This study confirms that a minimum ablative margin of 5 mm or larger from an AI-based autosegmentation/DIR method is associated with a very low risk of local disease progression in both the institution where the method was developed (0%) and three other institutions (less than 2%), supporting a minimum ablative margin of 5 mm as a threshold for A0 ablation. This study additionally identifies a margin of 3.2 mm measured using intraprocedural CECT as sufficient for A0 ablation. Moreover, a minimum ablative margins larger than 4.6 mm using intraprocedural CECT are associated with a negligible (less than 1%) reduction in local disease progression risk for each additional millimetre of ablative margin. Thus, considering the potential for complications due to additional ablation probe punctures and tissue damage by ablation, re-ablation of CRLMs with intraprocedurally confirmed minimum ablative margins larger than 3.2 mm should be considered carefully and avoided with minimum ablative margins larger than 4.6 mm. Previous studies reported required margins of 10 mm or larger when less accurate measurement techniques were used^12,14,30. This discrepancy between the present study and the previous studies is likely because of measurement error inherent in the other studies’ measurement techniques. Other studies using specialized ablation confirmation software on small or single-institution data sets identified margins in a similar range of 2–3 mm^6,10,11.

Different software-based ablation confirmation methods likely have different thresholds for A0 mainly due to differences in measurement accuracy. In addition, recent studies reporting very low (2–3 mm) minimum ablative margins for A0 ablation^10,11 had relatively small sample sizes when considering the historically low rate of sufficient margins (33% of CRLMs in the present study had a minimum ablative margin that was 5 mm or larger) combined with the rarity of local disease progression among patients who achieved sufficient margins during ablation (less than 2% in the present study). While the present study confirmed 5 mm as a good threshold for A0 margins using the author‘s ablation confirmation method, additional studies with large and diverse samples with long-term follow-up are warranted for other software packages using different imaging modalities and using different segmentation and registration techniques. In addition, as image resolution, quality improve, segmentation and registration techniques become more accurate, thresholds for A0 margins might further decrease.

The differences in patient populations regarding age, tumour size, extrahepatic disease, and rate of new CRLMs between the institutions in this study suggest significant differences in underlying tumour biology and referral patterns. Institution A is a quaternary academic cancer centre and therefore treats more patients with more aggressive disease. The other institutions are tertiary academic centres, with institutions B and D having similar intrahepatic outcomes. Institution C has a well-established stereotactic thermal ablation programme with a referral pattern that prefers ablation for resectable and ablateable CRLMs patients. Despite similar European and American guidelines for referral to thermal ablation, this study shows that large differences still exist and pose a significant barrier to comparing study outcomes after thermal ablation of CRLMs between centres. Further standardization of such referral patterns through multi-institutional studies and guidelines would improve the generalizability of study outcomes. Despite the differences in patient populations, the ablation procedure itself, and local and intrahepatic tumour progression rates, all four centres were similar with respect to OS.

The most frequent reason for exclusion in the present analysis was suboptimal imaging quality for image processing (11% of procedures), as it can lead to inaccurate segmentations and registrations—the main contributors to the overall accuracy of the minimum ablative margin measurement. An additional 4% of procedures were excluded due to inaccurate registrations. Prior series using distinct ablation confirmation methodologies have shown exclusion rates ranging from 30% to 52% due to suboptimal imaging quality and subsequent segmentation and registration issues^10,11,31, underscoring the strength of the authors’ ablation confirmation methodology across heterogeneous CT imaging techniques. More importantly, the authors’ methodology demonstrates the need for standardization of imaging techniques, especially regarding timing, contrast medium use, radiation dose, and the need for robust ablation confirmation methodologies suitable for use with varied CT imaging techniques. The present study also showed that a minimum ablative margin obtained using intraprocedural CECT better predicted 3-year local disease progression than a minimum ablative margin obtained using periprocedural CECT—confirming findings from previous studies^32,33. Thus, such standardization of the imaging modality guidance and technique during percutaneous thermal ablation procedures may lead to more reproducible and accurate minimum ablative margin measurements.

This study has limitations. First, the institutional data sets had important differences: data sets B, C, and D were smaller than data set A; and the baseline characteristics and incidences of new CRLMs differed significantly. Although the magnitude of the differences was surprising, inclusion of institutions with suspected differences in referral and ablation procedures was intentional to have a broad and diverse patient population. Second, the ablation confirmation software in this study reports a minimum ablative margin smaller than 0 mm as being a minimum ablative margin of 0 mm. Related to this, institution C had twice as many 0-mm minimum ablative margins as institution A, but less than half as many local disease progression events with 0-mm minimum ablative margins as institution A. With the use of stereotactic guidance, consistent intraprocedural imaging, and image fusion throughout, a 0-mm minimum ablative margin at institution C might represent complete tumour coverage without any ablative margin (0-mm minimum ablative margin). In contrast, 0-mm minimum ablative margins at institution A conceivably included more ablations where parts of the tumour were missed. Although this nuance was not captured by the authors’ software, it has little clinical relevance, as incompletely ablated tumours require immediate re-ablation, regardless of the amount of incomplete coverage. Third, although this was a multi-institution study, all patients were from high-volume academic centres. In addition, image processing was performed centrally at institution A and therefore the results might be different if the data were processed locally by users with less training on the software. Fourth, institution D used ultrasonographic guidance, so the last diagnostic CECT image and first follow-up image had to be used for margin quantification, which resulted in many 0-mm minimum ablative margins. Because institution D routinely performed contrast-enhanced ultrasonography 1 week before ablation to confirm the tumour size, 0-mm minimum ablative margins were more likely due to ablation zone shrinkage than tumour growth in the weeks before the ablation procedure, consistent with previous studies^32,33. Fifth, the main statistical analysis was performed on a per-tumour level, which might be affected by clustering. However, the sensitivity analysis performed using a mixed-effects model showed similar results and therefore any clustering effects are most likely minimal.

In conclusion, AI autosegmentation- and biomechanical DIR-based MAM is highly predictive of LDP after CRLM ablation. A MAM of 5 mm or larger confirmed by DIR demonstrated optimal local oncologic outcomes, with intraprocedurally confirmed MAM of more than 4.6 mm providing negligible improvements in LDP rates. We propose that an intraprocedural MAM of 5 mm or larger, confirmed using biomechanical DIR, serves as the A0 for CRLM thermal ablation.

Funding

Research reported in this publication was supported in part by the National Institutes of Health–National Cancer Institute (R01CA235564, R01CA221971, P30CA016672) and a donation by the Apache Corporation via the Image Guided Cancer Therapy Research Program at The University of Texas MD Anderson Cancer Center. This research was further supported by funding from the Propter Homines Stiftung, Vaduz, Liechtenstein and a GFI grant (CALM_GFI_22_01_F). I.P. was supported by a Postdoc.Mobility Fellowship from the Swiss National Science Foundation (P2BEP3_195444, P500PM_210871). AS was supported by an Image Guided Cancer Therapy (IGCT) T32 Training Program Fellowship from T32CA261856.

Author contributions

Iwan Paolucci (Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Validation, Writing—original draft, Writing—review & editing), Jessica Albuquerque Marques Silva (Data curation, Investigation, Writing—review & editing), Yuan-Mao Lin (Data curation, Investigation, Writing—review & editing), Gregor Laimer (Data curation, Writing—review & editing), Valentina Cignini (Data curation, Writing—review & editing), Francesca Menchini (Data curation, Writing—review & editing), Marcio Meira (Data curation, Writing—review & editing), Alexander Shieh (Data curation, Writing—review & editing), Caleb O’Connor (Resources, Software, Writing—review & editing), Kyle A. Jones (Resources, Software, Writing—review & editing), Carlo Gazzera (Funding acquisition, Writing—review & editing), Paolo Fonio (Funding acquisition, Writing—review & editing), Kristy K. Brock (Funding acquisition, Resources, Software, Supervision, Writing—review & editing), Marco Calandri (Data curation, Funding acquisition, Supervision, Writing—review & editing), Marcos Menezes (Funding acquisition, Supervision, Writing—review & editing), Reto Bale (Funding acquisition, Supervision, Writing—review & editing), and Bruno C. Odisio (Conceptualization, Funding acquisition, Project administration, Supervision, Writing—review & editing)

Disclosure

K.K.B.: grants from the National Institutes of Health and RaySearch Laboratories; a licensing agreement with RaySearch Laboratories; travel support from the American Association of Physicists in Medicine; patents planned, issued, or pending; and on an advisory board for RaySearch Laboratories. M.C.: speaker and consultant for Medtronic, Sanofi, AstraZeneca, and MedItalia. R.B.: research grants from Siemens Healthineers. B.C.O.: research grants from National Institutes of Health, Siemens Healthineers, and Johnson & Johnson; and consulting fees from Siemens Healthineers. The authors declare no other conflict of interest.

Supplementary material

Supplementary material is available at BJS online.

Data availability

Data generated or analysed during the study are available from the corresponding author by request.