Evaluating the early therapeutic response of radiofrequency ablation in rabbit VX2 lung tumors: a comparison of four diffusion weighted imaging models

Introduction

Tumor ablation techniques destroy tumor cells via either thermal energy [e.g., radiofrequency ablation (RFA) and microwave ablation (MWA)] or freezing (cryoablation), resulting in controlled coagulative necrosis (1). As minimally invasive interventions, their clinical utility ranges from providing curative-intent therapy for early-stage primary tumors and oligometastases to palliative local control for non-surgical candidates, thereby preserving lung parenchyma and accelerating patient recovery.

A meta-analysis has established that percutaneous ablation yields outcomes comparable to surgical resection (lobectomy or sublobectomy) in stage I non‑small cell lung cancer, with no significant differences in 1‑ to 5‑year overall, progression‑free, or cancerspecific survival (2). Specifically, MWA achieves a prognosis on par with lobectomy (2), while RFA shows survival outcomes akin to sublobectomy (3). Further evidence indicates that percutaneous ablation is noninferior to either surgery or stereotactic body radiotherapy (SBRT) (4,5). Critically, unlike radical resection or SBRT, local ablation has minimal impact on pulmonary function. However, the inherently low electrical conductivity and high thermal insulation of aerated lung tissue often result in incomplete marginal ablation, leading to local recurrence risk of as high as 30% (6). Consequently, the early detection of residual or recurrent disease is critical to guide timely re‑intervention and mitigate the risk of recurrence.

Following ablation, pulmonary tumors undergo diverse and slowly evolving morphological changes, such as fibrosis, nodule formation, cavitation, or atelectasis. However, these features are nonspecific and do not reliably indicate local tumor progression (7). Furthermore, proximity to major blood vessels or bronchi is a recognized risk factor for incomplete ablation, as the flow of blood or air at body temperature can dissipate accumulated heat, also known as “heat sink effect”. Clinically, definitive efficacy assessment is typically delayed until 4–6 weeks post-procedure to allow acute edema to subside. However, this prolonged waiting period often precludes timely re-intervention for residual tumor cells.

In parallel, although a ground-glass opacity (GGO) margin greater than 5 mm on immediate post-ablation imaging is associated with reduced local tumor progression (8-10), this finding remains non-specific, as it may represent either coagulation necrosis or benign inflammatory changes (11). This ambiguity potentially leads to an overestimation of the treatment effect and increased risk of recurrence. Therefore, identifying sensitive and accurate biomarkers within the ultra-early phase (specifically 1–3 days post-RFA) is critical for providing immediate feedback, potentially enabling supplementary ablation before patient discharge.

Beyond conventional imaging, diffusion-weighted imaging (DWI) has been investigated for the early detection of residual tumor or progression. The apparent diffusion coefficient (ADC) derived from the mono-exponential model has shown promise in identifying tumor progression prior to structural changes (12,13). However, prior studies are limited by the simplicity of this model and a lack of histopathological validation. More advanced DWI models, which enable the decoupling of water diffusion and microperfusion effects, remain largely unexplored in this setting, despite their potential for superior diagnostic accuracy.

Consequently, our primary objective was to evaluate and compare the ability of parameters from mono-exponential DWI (ADC), bi-exponential intravoxel incoherent motion (IVIM), the stretched exponential model (SEM), and diffusion kurtosis imaging (DKI) in differentiating between complete and incomplete ablation. These assessments were performed at critical early post-RFA time points, with histopathology as standard reference. We present this article in accordance with the ARRIVE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1379/rc).

Methods

Rabbits

All animal experiments were performed under a project license (No. JGLL-20181027-01) granted by institutional ethics board of Shanghai Public Health Clinical Center, in compliance with Shanghai Public Health Clinical Center institutional guidelines for the care and use of animals. A protocol was prepared before the study without registration. A total of 80 New Zealand White rabbits (provided by Shanghai Jiagan Biological Technology Co., Ltd.; permit number: SCXK2010-0028) bearing implanted VX2 lung tumors underwent RFA. The rabbits (40 males and 40 females) weighed 2.0–3.0 kg and were aged 3–5 months. Nine animals died during the procedure due to anesthetic accidents or complications. The remaining 71 rabbits were randomly assigned to four follow-up groups according to post-RFA time points: day 1 (n=18), day 3 (n=17), day 7 (n=18), and day 14 (n=18).

Subsequent attrition during follow-up included: two deaths and four technical failure including 1 for mono-exponential DWI, 2 for SEM and 1 for DKI on day 1; four deaths on day 3 due to delayed pneumothorax, abdominal distension, and diarrhea; four deaths on day 7 resulting from abdominal distension, diarrhea, and metastasis, along with one incomplete IVIM examination; and seven deaths on day 14 due to similar complications. As a result, 54 rabbits were included in the final analysis, distributed as follows: day 1 (n=16), day 3 (n=13), day 7 (n=14), and day 14 (n=11).

All rabbits were pathologically classified into a complete or an incomplete group. The overall study enrollment and grouping process are summarized in Figure 1.

Figure 1 Flow chart of lung transplanted VX2 tumor post-RFA and grouping. DKI, diffusion kurtosis imaging; IVIM, intravoxel incoherent motion; RFA, radiofrequency ablation.

Magnetic resonance imaging (MRI) scans

All MRI examinations were performed on a 1.5‑T clinical scanner (INTERA, Philips Healthcare, Best, The Netherlands) using a dedicated eight‑channel phased‑array coil (Model: CG‑RBC18‑H150‑AP, Chenguang Medical Technology Co., Ltd., Shanghai, China). Prior to scanning, rabbits were anesthetized via the marginal ear vein by intravenous injection of 3% pentobarbital sodium (1 mL/kg; Dingguo Biotechnology Co., Ltd., Shanghai, China).

Initial transverse T1‑ and T2‑weighted images (T1WI and T2WI) were acquired using fast spin‑echo (FSE) sequences with the following parameters: repetition time (TR)/echo time (TE) =577/10 ms, field of view (FOV) =160–180 mm × 160–180 mm, matrix =268×232, slice thickness =3 mm, gap =0.3 mm, number of signal averages (NSA) =3, sensitivity encoding (SENSE) factor =2, and 22 sections.

IVIM‑DWI was performed using a single‑shot echo‑planar imaging (SS‑EPI) sequence with 10 b‑values (0, 10, 20, 50, 100, 150, 200, 500, 800, and 1,000 s/mm²). Key parameters included: spectral presaturation with inversion recovery (SPIR), flip angle =90°, EPI factor =63, FOV = 180 mm × 180 mm, matrix =120×120, slice thickness =3 mm, gap =0.3 mm, TR/TE = 4716/107 ms, NSA =3, and 22 sections.

For DKI, four b‑values (0, 500, 1,000, and 2,000 s/mm²) were applied with TR/TE =6,484/119 ms and NSA =6, while all other geometric parameters matched the IVIM protocol.

The total acquisition time was 9 minutes 34 seconds for the combined T1WI, T2WI, and IVIM‑DWI, and 16 minutes 19 seconds for T1WI, T2WI, and DKI.

Image preprocessing

All DWI data were processed using FMRIB’s Diffusion Toolbox (FSL version 5.0; Oxford, UK; https://www.ndcn.ox.ac.uk/divisions/fmrib/analysis). Four diffusion weighted imaging models (14,15) were implemented as follows:

• Mono-exponential model: used to calculate the ADC value: $S_b/S_0=\exp \left(-b\times ADC\right)$, where S_b represents the single intensity at a given b value, and S₀ is the single intensity at b=0 s/mm². Two ADC values were derived: ADC₁ (b = 0, 800 s/mm²) and ADC₂ (b=0, 500 s/mm²).
• IVIM model: a bi-exponential model was employed to calculate the true diffusion coefficient (D), pseudo-diffusion coefficient (D*), and perfusion fraction (ƒ): $S_b/S_0=f\times \exp \left(-b\times D*\right)+\left(1-f\right)\times \exp \left(-b\times D\right)$. The product ƒD* reflects relative perfusion, calculated as the product of ƒ and D*.
• DKI model: applied to estimate the diffusion coefficient (D_k) and kurtosis (K): $S_b/S_0=\exp \left(-b\times D_k+1/6\times b^2\times D_k^2\times K\right)$.
• SEM model: used to compute the distributed diffusion coefficient (DDC) and the stretching exponent α: $S_b/S_0=\exp \left(-{\left(-b\times DDC\right)}^{\alpha }\right)$, where α represents the heterogeneity index.

Regions of interest (ROIs) were analyzed using the software of Image J (for mac, 1.51a, Wayne Rasband, National Institute of Health, USA). Two radiologists (Z.C. and F.S., with 10 and 15 years of experience in thoracic imaging, respectively) independently and blindly identified each lesion on mono-exponential DWI images with reference to T1WI and T2WI. They delineated the tumor contour at its largest cross-section to define an ROI, carefully avoiding pre-existing necrotic or cystic regions, blood vessels, and airways on the DWI image at b=500 s/mm². The same ROI was then applied to all other DWI model parametric maps.

Histopathological analysis

Following RFA and MRI at each designated time point (days 1, 3, 7 and 14), rabbits were euthanized via pentobarbital sodium overdose. Lung specimens were promptly harvested using the Heitzman method (inflation-fixation to preserve lung structure) and processed for hematoxylin and eosin (H&E) staining, CD31 immunohistochemistry, and TUNEL assay. Tissue sections (5 µm) were prepared from paraffin-embedded blocks and stained with H&E for histological evaluation.

For CD31 immunohistochemistry, 5 µm-thick paraffin sections were deparaffinized in xylene and stained via the peroxidase-antiperoxidase (PAP) method. After incubation with primary and secondary CD31 antibodies (ARG52748, Arigo Biolaboratories Corp., Taiwan, China), immunoreactivity was developed using 3,3'-diaminobenzidine (DAB, ZLI-9019, ZSJQ Bio-technology, Beijing, China), followed by hematoxylin counterstaining. A microvessel was defined as any CD31-positive endothelial cell or cell cluster, with or without a lumen, clearly separated from adjacent vessels or stromal structures. Microvessel density (MVD) was assessed by counting microvessels in five randomly selected high-power fields (200×) within vascular “hot spots”.

TUNEL staining was conducted using an in-situ cell death detection kit with fluorescein (No. 11684795910, Roche, Penzberg, Germany) following the manufacturer’s modified protocol (Version 11, April 2006) to evaluate cellular viability at the molecular level. Under fluorescence microscopy, green fluorescence indicated apoptotic or dead cells with deoxyribonucleic acid (DNA) fragmentation, while viable cells exhibited blue fluorescence. Extensive DNA fragmentation resulting from coagulative necrosis within the ablation zone was regarded as a true positive signal confirming successful cellular inactivation.

The ablation efficacy was defined as follows: complete ablation was defined as the complete absence of viable tumor cells within the ablation zone, characterized by varying degrees of coagulative necrosis, nuclear pyknosis, and karyorrhexis. Incomplete ablation was defined as the presence of any residual viable tumor cell nests, typically found at the periphery of the ablation zone, showing intact cellular structure and nuclear staining.

All stained sections were scanned using a digital panoramic imaging system (Pannoramic-250, 3D HISTECH Ltd., Budapest, Hungary) at 200× magnification and subsequently evaluated by a pathologist with over 30 years of experience in the field.

Statistical analysis

Statistical analyses were conducted using STATA (v14.0, USA), MedCalc (v17.1, Belgium), and GraphPad Prism (v6.05, USA). Continuous variables following a normal distribution are expressed as mean ± standard deviation (SD), whereas those with a non-normal distribution are summarized as median and interquartile range (IQR). Inter-observer agreement was evaluated using Bland-Altman analysis along with paired t-tests, with a coefficient of variation (CV).

Comparisons of multi-b-value DWI parameters between pre- and post-RFA scans, as well as between the complete and incomplete ablation groups, were performed using the independent t-test (for normally distributed data) or the Wilcoxon rank-sum test (for non-normally distributed data). Differences across multiple time points were assessed with the Kruskal-Wallis test (α=0.05). When statistical significance (P<0.05) was detected, post-hoc inter-group comparisons were conducted.

The diagnostic performance of each parameter in identifying complete ablation was evaluated using receiver operating characteristic (ROC) curve analysis, with results quantified by the area under the curve (AUC), sensitivity, specificity, and positive likelihood ratio (LR+). Correlations between changes in DWI parameters and MVD were assessed using Pearson (normal distribution) or Spearman (non-normal distribution) correlation analysis. All results are presented with 95% confidence intervals (CI), and a P value <0.05 was considered statistically significant.

Results

Reproducibility of DWI parameters

Table 1 summarizes the DWI parameters derived from the four models at each postoperative time point. Inter-individual reproducibility was moderate for ADC₂ (CV: 13.25–20.26%), DDC (CV: 10.65–13.86%), α (CV: 11.21–14.69%), and D_k (CV: 19.00–23.88%). In contrast, poor reproducibility was observed for ADC₁ (CV: 20.79–29.60%), K (CV: 25.49–30.43%), and all IVIM-derived parameters (D, D*, and ƒ), with CVs ranging from 25.25% to 178%.

No statistically significant differences were found in interobserver measurements (all P>0.05). As shown in Table S1, interobserver reproducibility at baseline was good for parameters from the mono-exponential, SEM, and DKI models, with CVs of 6.15% for ADC₁, 2.18% for ADC₂, 3.33% for DDC, 3.45% for α, 7.18% for D_k, and 8.65% for K. In comparison, IVIM-derived parameters showed only moderate reproducibility, with CVs of 18.62% for D, 15.65% for D*, and 12.12% for ƒ.

Pre- and post-RFA DWI parameter changes

As demonstrated in Table 2 and Figures S1,S2, at early time points (days 1 and 3), significant increases from baseline were observed in ΔADC₁ [(0.35±0.37)×10⁻³ mm²/s, P=0.003; (0.33±0.33)×10⁻³ mm²/s, P=0.004], ΔADC₂ [(0.57±0.35)×10⁻³ mm²/s, P<0.001; (0.41±0.43)×10⁻³ mm²/s, P=0.005], ΔD [(0.39±0.46)×10⁻³ mm²/s, P=0.004; (0.19±0.24)×10⁻³ mm²/s, P=0.01], Δƒ (8.71±13.83, P=0.02; 11.43±16.71, P=0.03), ΔDDC [(0.24±0.30)×10⁻³ mm²/s, P=0.009; (0.25±0.33)×10⁻³ mm²/s, P=0.02], Δα (0.08±0.10, P=0.009; 0.09±0.11, P=0.02), ΔD_k [(0.26±0.39)×10⁻³ mm²/s, P=0.02; (0.42±0.42)×10⁻³ mm²/s, P=0.004]. A significant increase in ΔK was also noted on day 3 (0.33±0.14, P=0.03), though no significant difference was found on day 1 (0.39±0.08, P=0.39). In contrast, ΔD* and ΔƒD* showed no significant changes (all P>0.05), suggesting limited utility in reflecting early treatment response.

At later time points (days 7 and 14), significant elevations from baseline were maintained in ΔADC₂ [(0.45±0.52)×10⁻³ mm²/s, P=0.007; (0.61±0.40)×10⁻³ mm²/s, P<0.001], ΔD [(0.39±0.67)×10⁻³ mm²/s, P=0.03; (0.54±0.45)×10⁻³ mm²/s, P=0.002), ΔDDC [(0.29±0.31)×10⁻³ mm²/s, P=0.004; (0.37±0.22)×10⁻³ mm²/s, P<0.001], Δα (0.10±0.10, P=0.004; 0.13±0.08, P<0.001), ΔD_k [(0.60±0.56)×10⁻³ mm²/s, P=0.002; (0.35±0.38)×10⁻³ mm²/s, P=0.01]. ΔADC₁ was significantly elevated on day 14 [(0.23±0.25)×10⁻³ mm²/s, P=0.01], with no difference on day 7 [(−0.02±0.31)×10⁻³ mm²/s, P=0.82]. Conversely, ΔK decreased significantly on day 7 (−0.35±0.40, P=0.006) and showed no significant change on day 14 (−0.15±0.29, P=0.11). No significant differences were observed in ΔD*, Δƒ, and ΔƒD* at these time points (all P>0.05).

Diagnostic performance of fitting models

As shown in Table 3 and Figure 2, DKI-derived ∆D_k was significantly higher in the complete ablation group than in the incomplete ablation group at all early and late time points [day 1: (0.71 vs. 0.07)×10⁻³ mm²/s, P<0.001; day 3: (0.72 vs. 0.07)×10⁻³ mm²/s, P=0.002; day 7: (0.52 vs. 0.03)×10⁻³ mm²/s, P=0.006; day 14: (0.43 vs. 0.07)×10⁻³ mm²/s, P=0.02]. Additionally, on day 3, the complete ablation group showed a significantly greater increase in ∆ADC₂ [(0.59 vs. 0.14)×10⁻³ mm²/s; P=0.008] and a more pronounced decrease in ∆K (−0.37 vs. −0.05; P=0.005). During the later phase (days 7–14), the complete ablation group presented significantly elevated ∆D [(0.52 vs. −0.42)×10⁻³ mm²/s; P=0.007), but reduced ∆D* [(−2.95 vs. 4.54)×10⁻³ mm²/s; P=0.01) and ∆ƒD* [(−1.56 vs. 4.19)×10⁻³ mm²/s; P=0.01], along with higher ∆DDC [(0.37 vs. 0.05)×10⁻³ mm²/s; P=0.03] and ∆α (0.13 vs. 0.02; P=0.03) on Day 7. By Day 14, the incomplete ablation group was characterized by significantly increased ∆D* [(3.58 vs. −3.63)×10⁻³ mm²/s; P=0.049] and ∆ƒD* [(1.27 vs. −2.11)×10⁻³ mm²/s; P=0.049]. However, ∆ADC₁ and ∆ƒ were limited in differentiating complete and incomplete ablation.

Figure 2 Box plots showing the comparison of diffusion-weighted imaging parameters between complete (tumor-free) and incomplete (residual tumor) ablation groups in the rabbit VX2 lung tumor models at four-times-points (days 1, 3, 7 and 14) post-ablation. The parameters measured include the ΔADC at two b values: (A) ΔADC1 (b=0, 800 s/mm²), (B) ΔADC2 (b=0, 500 s/mm²); IVIM parameters: (C) ΔD, (D) ΔD*, (E) Δƒ, (F) ΔƒD*; SEM: (G) ΔDDC, (H) Δα; and DKI parameters: (I) ΔDk, (J) ΔK. Δ, change (post − pre); α, stretching exponent; ADC, apparent diffusion coefficient; D, true diffusion coefficient; D*, pseudo-diffusion coefficient; DDC, distributed diffusion coefficient; Dk, the DKI-derived diffusion coefficient; DKI, diffusion kurtosis imaging; ƒ, perfusion fraction; ƒD*, relative perfusion (ƒ×D*); IVIM, intravoxel incoherent motion; K, kurtosis; RFA, radiofrequency ablation; SEM, stretched exponential model.

As shown in Table 4 and Figure 3, on day 1 post-RFA, the AUC of the ∆D_k value for assessing the completeness of tumor ablation was 1.00. On day 3, the AUCs of ∆ADC₂, ∆D_k, and ∆K for evaluating ablation completeness were 0.93, 0.98, and 0.95, respectively, with no statistically significant inter-comparison differences (all pairwise comparisons >0.02; α=0.05/3). On Day 7, the AUCs of ∆D, ∆D*, ∆ƒD*, ∆DDC, ∆D_k and ∆α for determining ablation completeness were 1.00, 0.95, 0.95, 0.85, 0.85, and 1.00, respectively, and no significant differences were observed among the six parameters (all pairwise comparisons >0.01; α=0.05/15). On day 14, the AUCs of ∆D*, ∆ƒD* and ∆D_k were 0.88, 0.88, and 1.00, respectively, again showing no significant differences (all pairwise comparisons >0.02; α=0.05/3). Figures 4-6 demonstrate post-RFA cases evaluated with the multi-b-value DWI models.

Figure 3 Receiver operating characteristic curves of diffusion-weighted imaging parameters in assessing ablative efficacy. Data are shown for post-RFA days 1 (A), 3 (B), 7 (C), and 14 (D). α, the stretching exponent; ADC, apparent diffusion coefficient; D, true diffusion coefficient; DDC, distributed diffusion coefficient; Dk, the DKI-derived diffusion coefficient; DKI, diffusion kurtosis imaging; Dstar, pseudo-diffusion coefficient; ƒ, perfusion fraction. ƒDstar, relative perfusion (ƒ ×D*); K, kurtosis; RFA, radiofrequency ablation.

Figure 4 Correlation between IVIM parametric maps (yellow outlines indicate ROIs) and histopathology in a rabbit VX2 lung tumor model following RFA (day 7). (A,B) Parametric maps of D before and after RFA (ROI values: 0.64×10⁻³vs. 1.96×10⁻³ mm²/s). (C,D) Parametric maps of D* before and after RFA (ROI values: 74.97×10⁻³vs. 74.42×10⁻³ mm²/s). (E,F) Parametric maps of ƒ before and after RFA (ROI values: 38.05% vs. 44.49%). (G) H&E staining of the post-ablation tumor region (yellow circle). (H) TUNEL staining showing extensive tumor cell devitalization (indicated by green fluorescence). D, true diffusion coefficient; D*, pseudo-diffusion coefficient; ƒ, perfusion fraction; H&E, hematoxylin and eosin; IVIM, intravoxel incoherent motion; RFA, radiofrequency ablation; ROI, region of interest.

Figure 5 Correlation between SEM-DWI parametric maps (yellow outlines indicate ROIs) and histopathology in a rabbit VX2 lung tumor following complete RFA (day 7). (A,B) DDC parametric maps before and after RFA (ROI values: 1.76×10⁻³vs. 2.10×10⁻³ mm²/s). (C,D) α parametric maps before and after RFA (ROI values: 0.57 vs. 0.69). (E) Inflated and fixed gross specimen showing extensive tumor destruction with ill-defined margins (yellow circle) post-RFA. (F) TUNEL staining of the corresponding section; the yellow circle highlights widespread apoptosis and tumor cell death. α, the stretching exponent; DDC, distributed diffusion coefficient; DWI, diffusion-weighted imaging; RFA, radiofrequency ablation; ROI, region of interest; SEM, stretched exponential model.

Figure 6 Correlation between DKI parametric maps (yellow outlines indicate ROIs) and histopathology in a rabbit VX2 lung tumor following RFA (day 14). (A,B) Mean Dk parametric maps before and after RFA (ROI values: 1.53×10⁻³vs. 1.69×10⁻³ mm²/s). (C,D) Mean K parametric maps before and after RFA (ROI values: 1.21 vs. 0.88). (E) Inflated and fixed gross specimen showing peripheral coagulative necrosis and residual tumor with a ‘fish-flesh’ appearance (yellow circle) near the hilum. (F,G) TUNEL staining of the corresponding section (F: ×200); tissue near hilum (G, yellow arrows in circle) indicating residual vital cells (F, yellow arrows and red triangles). Dk, the DKI-derived diffusion coefficient; DKI, diffusion kurtosis imaging; K, kurtosis; RFA, radiofrequency ablation; ROI, region of interest.

Correlations of DWI parameters and MVD

As presented in Table S2, ∆D* was positively correlated on Day 1 (r=0.559, P=0.03) and negatively on day 3 (r=−0.556, P=0.04). ΔD_k showed a strong negative correlation on day 14 (r=−0.840, P=0.003). Trends toward significance were noted for ∆ƒD* on day 1 (r=0.490, P=0.055) and ΔD on day 14 (r=−0.555, P=0.06). No significant correlations were observed for ΔADC₁, ΔADC₂, or ΔDDC across time points.

Discussion

This study demonstrates the superior diagnostic value of diffusion-weighted parameters derived from DWI, IVIM, SEM, and DKI models in differentiating complete from incomplete ablation. Notably, the DKI-derived parameter ∆D_k emerged as a robust discriminator, showing significant differences in both early and late post-RFA phases and maintaining a strong negative correlation with MVD at day 14. This early indication was further corroborated on day 3 by a concurrent reduction in ∆K and an increase in ∆ADC₂ (b=0, 500 s/mm²). In the later phase (day 14), significantly elevated IVIM-derived ∆D* and ∆ƒD* values in the incomplete group possibly indicated the presence of residual tumor activity and compensatory neoangiogenesis.

RFA induces coagulative necrosis and structural disintegration of tumor cells via instantaneous high temperatures, thereby resulting in the expansion of the extracellular space and a subsequent reduction in water diffusion restriction (16). Consistent with previous studies reporting significantly elevated ADC values post-ablation (12,16), our results demonstrated a notable increase in tumor ADC. This trend was further refined by concurrent rises in SEM-derived DDC, IVIM-derived D, and DKI-derived D_k values, which provide a more precise quantification of water mobility and intravoxel diffusion heterogeneity. Collectively, these findings confirm that quantitative DWI parameters can effectively reflect RFA treatment-induced cellular-level microenvironment alterations in tumors.

In evaluating the mono-exponential DWI model, this study utilized two sets of ADC values (acquired with b values of 0, 500 and 0, 800 s/mm², respectively) to assess tumor response following RFA. On post-ablation day 1, the increase in ADC was slightly lower in the complete ablation group than in the incomplete group. This finding may be explained by transient acute perfusion increases and inflammatory responses, as supported by a prior radiopathological correlation study (17) and our observation of mildly elevated microperfusion parameters (D*, ƒ and ƒD* values) on post-RFA day 1. These findings collectively indicate that ADC measurements obtained at 24 hours post-RFA may be suboptimal for early treatment evaluation, as they are susceptible to interference from acute peritumoral changes. Notably, this observation contrasts with a recent study by Vogl et al., which reported that ADC values (using b=0, 50, 400, 800 s/mm²) could predict efficacy at 24 hours after MWAs (12). However, our findings indicate that the choice of b-value is critical; ADC₂ (b=0, 500 s/mm²) outperformed ADC₁ (b=0, 800 s/mm² in distinguishing outcomes as early as day 3. Furthermore, ADC₂ demonstrated excellent inter-observer agreement (CV%: 2.18% vs. 6.15% for ADC₁), likely attributable to enhanced image quality and a higher signal-to-noise ratio, despite potential confounding from T2 shine-through effects.

To overcome the diagnostic limitations of ADC in the immediate post-operative phase, the non-Gaussian DKI model was introduced. Our results demonstrated that D_k significantly outperformed ADC at this ultra-early stage, achieving excellent discrimination (AUC =1.00) as early as day 1. The superiority of D_k stems from the mathematical robustness of the DKI model in heterogeneous tissues. RFA induces a complex microenvironment characterized by coagulative necrosis, cell membrane disruption, and inflammatory edema, where water diffusion follows a non-Gaussian distribution. While ADC assumes a Gaussian distribution and may mask subtle microstructural variations, D_k explicitly corrects for the kurtosis effect. This allows D_k to capture the stark contrast between the intact membranes of residual tumor and the membrane lysis in necrosis more accurately.

While D_k serves as a robust indicator of diffusivity, the K value reflects tissue microstructural complexity. Previous studies have established DKI-derived K value (18,19) as a sensitive biomarker for evaluating hepatocellular carcinoma after RFA, where its decrease indicates successful treatment. Similarly, in neoadjuvant chemotherapy for breast cancer, K has demonstrated high accuracy in predicting treatment response, consistently outperforming traditional ADC and IVIM models (20,21). Nevertheless, its application in lung tumors following ablation remains largely unexplored. In this study, K values progressively declined after RFA, due to the loss of microstructural integrity. The significantly higher K values in the incomplete group reflect the inherent complexity of residual viable tumor. However, it should be noted that clinical translation of DKI faces challenges; specifically, acquiring high b-values in free-breathing patients may suffer from lower signal-to-noise ratios and respiratory artifacts compared to our anesthetized rabbit model, necessitating rapid acquisition protocols or respiratory gating for future clinical application.

In contrast to the early sensitivity of DKI, SEM and IVIM parameters served as later indicators of treatment response, becoming discernible from days 7 to post‑RFA. The inversion of the correlation between ∆D* and MVD from days 1 (positive) to 3 (negative) reflects the dynamic evolution of the post-ablation microenvironment. On day 1, acute reactive hyperemia dominates, where structural MVD aligns with functional perfusion (D*). By day 3, however, the relationship is likely confounded by vascular thrombosis and the “vascular lock-in” effect; while CD31 staining still identifies endothelial structures (MVD), these vessels may be functionally occluded, leading to a decoupling of structure and function. By post-RFA day 14, elevated D and ƒD* were suggestive of residual tumor tissue. This is consistent with findings by Cheng et al. (22), who reported persistent pulmonary flow in incomplete ablation zones. Similarly, Ma et al. discovered a phase of transient proliferation and increased MVD in residual pulmonary tumors after RFA (23).

Although the SEM has been used in prior studies (24,25) to detect early diffusion heterogeneity following chemotherapy, its utility in the RFA was relatively limited. In this study, SEM facilitated the differentiation of ablation outcomes on postoperative day 7, its diagnostic performance (AUC =0.88 for both DDC and α) was inferior to that of the DKI and IVIM models. By day 14, SEM parameters no longer reliably distinguished ablation completeness, likely due to overlapping pathological features such as residual tumor growth and connective tissue proliferation at the ablation margin. Consequently, we recommend a stage-dependent selection of DWI models: the DKI model is preferred for immediate post-ablation feedback (days 1–3) to detect early cellular necrosis, while the IVIM model is more suitable for monitoring vascular remodeling and angiogenesis during the subacute-to-late phase (days 7–14).

There were several limitations in this study. First, the sample size in each subgroup was limited, particularly for the incomplete ablation group at the 7- and 14-day follow-ups. As this was a preclinical animal study, these results require validation in clinical cohorts to confirm their generalizability. Second, parameter measurements may have been influenced by imaging artifacts stemming from the small tumor size in rabbit lungs, the proximity of lesions to major vessels and bronchi, and the inherent challenges of a free-breathing acquisition protocol. Third, although eddy-current correction was applied to all images prior to analysis, potential distortions in DWI could still have affected the precision of image-pathology co-registration. Finally, the moderate reproducibility observed for certain parameters highlights the need for further technical optimization in future investigations.

Conclusions

This study confirms that a multi-parametric diffusion-weighted MRI approach effectively evaluates RFA outcomes in the rabbit VX2 lung tumor model. Among the assessed parameters, D_k proved to be the most robust biomarker across all time points, while ADC and K values offered additional early predictive value. These findings underscore the potential of advanced DWI models to serve as a comprehensive, non-invasive tool for assessing the treatment efficacy of patients with pulmonary tumors.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the ARRIVE reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1379/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1379/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1379/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1379/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All animal experiments were performed under a project license (No. JGLL-20181027-01) granted by institutional ethics board of Shanghai Public Health Clinical Center, in compliance with Shanghai Public Health Clinical Center institutional guidelines for the care and use of animals.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Ahmed M, Brace CL, Lee FT Jr, et al. Principles of and advances in percutaneous ablation. Radiology 2011;258:351-69. [Crossref] [PubMed]
2. Chan MV, Huo YR, Cao C, et al. Survival outcomes for surgical resection versus CT-guided percutaneous ablation for stage I non-small cell lung cancer (NSCLC): a systematic review and meta-analysis. Eur Radiol 2021;31:5421-33. [Crossref] [PubMed]
3. Iguchi T, Hiraki T, Matsui Y, et al. Survival Outcomes of Treatment with Radiofrequency Ablation, Stereotactic Body Radiotherapy, or Sublobar Resection for Patients with Clinical Stage I Non-Small-Cell Lung Cancer: A Single-Center Evaluation. J Vasc Interv Radiol 2020;31:1044-51. [Crossref] [PubMed]
4. Lam A, Yoshida EJ, Bui K, et al. A National Cancer Database Analysis of Radiofrequency Ablation versus Stereotactic Body Radiotherapy in Early-Stage Non-Small Cell Lung Cancer. J Vasc Interv Radiol 2018;29:1211-1217.e1. [Crossref] [PubMed]
5. Mendogni P, Daffrè E, Rosso L, et al. Percutaneous lung microwave ablation versus lung resection in high-risk patients. A monocentric experience. Acta Biomed 2020;91:e2020002. [Crossref] [PubMed]
6. Xu J, Xie Q, Ye X. Effectiveness and safety of percutaneous microwave ablation and radiofrequency ablation in the treatment of pulmonary metastasis: A systematic review and meta-analysis. J Cancer Res Ther 2025;21:804-9. [Crossref] [PubMed]
7. Vespro V, Bonanno MC, Andrisani MC, et al. CT after Lung Microwave Ablation: Normal Findings and Evolution Patterns of Treated Lesions. Tomography 2022;8:617-26. [Crossref] [PubMed]
8. Najafi A, de Baere T, Purenne E, et al. Risk factors for local tumor progression after RFA of pulmonary metastases: a matched case-control study. Eur Radiol 2021;31:5361-9. [Crossref] [PubMed]
9. Araujo-Filho JAB, Menezes RSAA, Horvat N, et al. Lung radiofrequency ablation: post-procedure imaging patterns and late follow-up. Eur J Radiol Open 2020;7:100276. [Crossref] [PubMed]
10. Liu BD, Ye X, Fan WJ, et al. Expert consensus on image-guided radiofrequency ablation of pulmonary tumors: 2018 edition. Thorac Cancer 2018;9:1194-208. [Crossref] [PubMed]
11. Chen J, Lin XN, Miao XH, et al. Evaluation of the correlation between infrared thermal imaging-magnetic resonance imaging-pathology of microwave ablation of lesions in rabbit lung tumors. J Cancer Res Ther 2020;16:1129-33. [Crossref] [PubMed]
12. Vogl TJ, Emara EH, Elhawash E, et al. Feasibility of diffusion-weighted magnetic resonance imaging in evaluation of early therapeutic response after CT-guided microwave ablation of inoperable lung neoplasms. Eur Radiol 2022;32:3288-96. [Crossref] [PubMed]
13. Zhang Y, Wei H, Song B. Magnetic resonance imaging for treatment response evaluation and prognostication of hepatocellular carcinoma after thermal ablation. Insights Imaging 2023;14:87. [Crossref] [PubMed]
14. Raju J, Shyna A, John A, et al. Transforming Medical Imaging with IVIM MR Imaging: A Comprehensive Review of Advances and Challenges in Perfusion and Diffusion Analysis. Applied Magnetic Resonance 2025;56:803-43.
15. Tang L, Zhou XJ. Diffusion MRI of cancer: From low to high b-values. J Magn Reson Imaging 2019;49:23-40. [Crossref] [PubMed]
16. Okuma T, Matsuoka T, Yamamoto A, et al. Assessment of early treatment response after CT-guided radiofrequency ablation of unresectable lung tumours by diffusion-weighted MRI: a pilot study. Br J Radiol 2009;82:989-94. [Crossref] [PubMed]
17. Yamamoto A, Nakamura K, Matsuoka T, et al. Radiofrequency ablation in a porcine lung model: correlation between CT and histopathologic findings. AJR Am J Roentgenol 2005;185:1299-306. [Crossref] [PubMed]
18. Yuan ZG, Wang ZY, Xia MY, et al. Comparison of diffusion kurtosis imaging versus diffusion weighted imaging in predicting the recurrence of early stage single nodules of hepatocellular carcinoma treated by radiofrequency ablation. Cancer Imaging 2019;19:30. [Crossref] [PubMed]
19. Zhang G, Li W, Wang G, et al. Multimode tumor ablation therapy induced different diffusion and microvasculature related parameters change on functional magnetic resonance imaging compared to radiofrequency ablation in liver tumor: An observational study. Medicine (Baltimore) 2020;99:e20795. [Crossref] [PubMed]
20. Gong X, Wang X, Wang L, et al. Comparing Multi-Value Diffusion MRI Models for Predicting Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast Cancer. Radiology 2025;316:e242969. [Crossref] [PubMed]
21. Chen X, Luo Y, Xie Z, et al. Prediction of neoadjuvant chemotherapy efficacy in breast cancer: integrating multimodal imaging and clinical features. BMC Med Imaging 2025;25:118. [Crossref] [PubMed]
22. Cheng Z, Wang Y, Yuan M, et al. CT perfusion imaging can detect residual lung tumor early after radiofrequency ablation: a preliminary animal study on both tumoral and peri-tumoral region assessment. J Thorac Dis 2022;14:64-75. [Crossref] [PubMed]
23. Ma L, Zhou N, Qi Y, et al. Changes of proliferation and angiogenesis of residual tumor in rabbit lung after radiofrequency ablation. Zhonghua Yi Xue Za Zhi. 2014;94:1671-3.
24. Suo S, Yin Y, Geng X, et al. Diffusion-weighted MRI for predicting pathologic response to neoadjuvant chemotherapy in breast cancer: evaluation with mono-, bi-, and stretched-exponential models. J Transl Med 2021;19:236. [Crossref] [PubMed]
25. Ren H, Chen X, Yang J, et al. Prediction of induction chemotherapy response in locoregionally advanced nasopharyngeal carcinoma based on three pretreatment non-Gaussian diffusion MRI models. BMC Med Imaging 2025;25:261. [Crossref] [PubMed]