Diffusion kurtosis imaging and conventional diffusion weighted imaging to assess electrochemotherapy response in locally advanced pancreatic cancer

Electrochemotherapy (ECT) is a promising tool for treatment of deep tumours.¹, ², ³, ⁴, ⁵, ⁶ This ablative technique combines the administration of chemotherapeutic drugs with electric pulses for cell membrane electroporation.⁷ In our previous studies⁶, ⁷, ⁸, ⁹, we demonstrated the safety and efficacy of the treatment in locally advanced pancreatic cancer. However, the correct assessment of the efficacy of ECT is a challenge for a radiologist since tumour necrosis is not always correlated with a dimensional change so that the response evaluation criteria in solid tumour (RECIST) are not adequate to assess the ECT treatment.⁹ In this scenario, we evaluated functional parameters extracted by diffusion weighted imaging (DWI).¹⁰ The practice of DWI into a standard magnetic resonance (MR) protocol is increasing, thanks to its ability to detect and characterize a lesion so as its capability in the assessment of treatment response.¹¹, ¹² Promising results for improved detection and monitoring of therapeutic effects, in terms of prediction and early response assessment, have been reported, amongst others, for the liver¹³, pancreas¹⁴, kidneys¹⁵ and prostate.¹⁶ Several authors have reported that DWI sequence have a potential in the evaluation of patients with locally advanced pancreatic cancer, especially improving the staging; nevertheless, it is still unclear whether and how DWI could be helpful for identification, characterization, prognostic stratification and follow-up during treatment. DWI assessment may be done qualitatively or quantitatively, with a mono-exponential (apparent diffusion coefficient map, ADC) or bi-exponential analysis (intravoxel incoherent motion model or diffusion kurtosis imaging, DKI) of data.¹⁷, ¹⁸, ¹⁹ DWI signal is due to the water mobility that reproduces indirectly tissue structures.¹⁴, ¹⁸ Traditionally DWI approach is based on the hypothesis that water molecules diffuse within a voxel following a single component direction according to a Gaussian behavior, without any restriction.¹⁸, ¹⁹ However, according to the presence of microstructures, random motion or diffusion of thermally agitated water molecules within biologic tissues exhibits non-Gaussian phenomena due to the presence of microstructures.¹⁹ This model, proposed by Jensen et al. in 2005, considers a non-Gaussian diffusion model called DKI.¹⁹ This approach evaluates the kurtosis coefficient (K) that shows the deviation of tissue diffusion from a Gaussian model, and the diffusion coefficient (D) with the correction of non-Gaussian bias. Several researches have shown that DKI is more performing than conventional ADC in tumour detecting and grading assessment.¹⁹, ²⁰, ²¹, ²², ²³ Kartalis et al.²⁴ reported that the correction for kurtosis effects has the potential to increase the diagnostic accuracy of DWI in patients with pancreatic ductal adenocarcinoma. To the best of our knowledge, no studies that evaluated the diagnostic performance of DKI parameters to assess ECT response have been published.

The purpose of this retrospective study is to evaluate the diagnostic performance of ADC by conventional mono-exponential approach and MD and MK by DKI to assess ECT response.

According to the results reported by our previous study⁹, among 13 patients subjected to MRI, 10 showed a significant response while 3 patients were classified in a sTable disease (Table 3).

ADC and MK did not show differences statistically significant between the values pre and post treatment (p value = 0.02 at Wilcoxon test) and between the percentage changes of responder and not responder patients (p value > 0.05 at Wilcoxon test). Only MD showed a significant variation of values between pre and post treatment (p value = 0.02 at Wilcoxon test) and a significant statistically difference for percentage changes between responders and not responders (p value = 0.01 at Kruskal Wallis test).

Figure 1 reports boxplot for each parameter pre and post treatment between responder and not responder patients. Figure 2 shows boxplot for percentage change (delta) between pre and post treatment for each parameters.

Figure 1

Figure 2

Table 4 reports the diagnostic performance of ADC and DKI derived parameters to assess ECT response.

Exclusively MD, among investigated parameters, had a good diagnostic performance with a sensitivity of 80%, a specificity of 100% (Table 1) and area under ROC of 0.933 (Figure 3).

Figure 3

Figure 4 shows a case of significant response on conventional MR images, ADC map and DKI derived parameters maps.

Figure 4

ECT treatment is a very promising tool in oncological patients. The establishment and expansion of ECT in deep-seated tumours (e.g., liver, bone metastases, pancreas) has opened new opportunities for minimally invasive treatment of metastases and carcinomas.⁷ ECT is usually applied as palliative treatment for patients with not resecTable lesion, causing an improvement of quality of life. Several researchers have evaluated it as a treatment of advanced neoplastic lesions in which radical surgery is not possible (e.g., due to lesion location, size, and/or number). ECT allows treatment of lesions in proximity of vital structures like vessels and nerves. The safety profile of ECT is favourable, with local and transient adverse events.⁷ Safety and efficacy of ECT have been showed in a prospective series of patients with unresecTable Perihilar-Cholangiocarcinoma (PHCCA).⁵ A prospective study conducted in 10 patients suffering from hepatocellular carcinomas (HCC) proved that ECT is a feasible and safe treatment. At average of 20.5 months after ECT 15 out of 17 lesions showed a complete response.²⁷ Recently, Boc et al. characterized ultra-sonographic (US) findings during and after electrochemotherapy of liver. US provides a tool for assessment of appropriate electrode insertion for intraoperative electrochemotherapy of liver tumours and assessment of the appropriate coverage of a tumour with a sufficiently strong electric field and can serve as predictor of the response of tumours.²⁸ As we reported in our previous studies, ECT can be used in locally advanced pancreatic cancer with no side effects or major complications to surrounding viscera that required medical or surgical treatment.⁷, ⁸ The major limit of our study is due to how we should assess the ablated area. In fact, we demonstrated in⁹ that morphological assessment based on RECIST criteria in not sufficient to stratify the patients as responders or not responders.⁹ Conversely, a functional assessment, based on the evaluation of parameters extracted by DCE-MRI and DWI, could be used to identify responders and not responders patients.⁹ In the current study, we evaluated conventional DWI parameters (ADC) and the parameters extracted by DKI (MD and MK) in the assessment of ECT treatment. Only MD by DKI showed a significant variation of value between pre and post treatment and a significant statistically difference for percentage change between responders and not. When we evaluated the diagnostic accuracy of ADC and DKI parameters, MD had a good accuracy with and area under ROC of 0.933. To the best of our knowledge, the few reports on Kurtosis in the assessment of pancreatic tumours, investigate the role of these functional parameters during the characterization of the lesions²⁹, other reports on the post treatment assessment by DKI, are not related to pancreas tumours.³⁰, ³¹, ^{32 33}

Diffusion is related to the random Brownian motion of water molecules. In a homogenous medium, diffusion is described by a Gaussian model, while in living tissues is related also by interactions with other molecules and cell membranes and is therefore thought to follow a non-Gaussian model. Also, ADC, in conventional DWI, is influenced by b-values acquired.¹⁰ DKI is an extension of diffusion tensor imaging that evaluates the microstructure features of tissues in a non-Gaussian model.¹⁹ Two quantitative parameters can be extracted by DKI model, including kurtosis values (K, representing deviation from a Gaussian distribution) and diffusion coefficient (D, defining as a corrected ADC for non-Gaussian bias).¹⁹ According to Jensen et al. DKI should exhibit a higher sensitivity than conventional DWI for cancer detection.¹⁹ Shen et al. assessed the accuracy of DKI parameters in evaluating malignant from benign lesions, performing a systematic meta-analysis.³⁴ They evaluated 14 studies from 2012 to 2018, for a total of 1847 lesions (895 benign and 952 malignant) in 1107 patients; only two studies were performed with a 1.5 Tesla MR system. The diagnostic performance of three quantitative data, K, D and ADC obtained from MRI with DKI and DWI, for differentiating malignant from benign lesions was evaluated. Pooled results indicated both K and D had a good or excellent diagnostic performance in separating malignant cancer from benign lesions, but D may be more superior because it had the higher AUC.³⁰ No study on pancreatic cancer was included in this meta-analysis.

Hu et al. evaluated DKI in assessing response in rectal cancer compared to conventional DWI.³³ They evaluated ADC, MD (mean diffusion), and MK (mean kurtosis) showing that the MK_pre and MK_post were much lower for the responder than for not responder patients. Also, the MD_post and the change ratio of MD increased, whereas there are no significant differences for ADC_pre, MD_pre and the change of MK ratio between responder and not responder patients. The MK_post had higher sensitivity and specificity compared to other data.³³

DWI has demonstrated to be a great promise as an imaging biomarker. However, it still suffers from inconsistencies in imaging acquisition parameters and DWI analysis that lead to substantial issues with reproducibility. Newer imaging techniques such as DKI may extend the potential of DWI by better characterizing and assessing the lesions pre and after treatment. Therefore, DKI should be added to the routine protocol. Several limitations in our study must be mentioned. First, since this was a retrospective study, there may have been potential selection bias. It will be necessary to perform larger studies to validate our findings. Second our data were acquired with a maximum b value of 1000 s/mm². In general, very high b-values are recommended for the evaluation of non-Gaussian kurtosis in brain applications.¹⁹ However, for abdominal studies, taking into account the lower signal to noise ratio (SNR) and the lower T2 relaxation times of the various organs compared to the brain, very high b-values are not usually applied. Recently, various authors have shown that kurtosis effects could be detecTable in abdominal and whole-body applications even when using maximum b-values of 800 s/mm² or less at 3T.³⁵, ³⁶, ³⁷ We applied multiple b-values with a maximum of 1000 s/mm² that, coupled with the use of a parallel imaging factor, resulted in images with accepTable SNR at 1.5T. Third, the size of our population is very small. In fact, this study had the only objective to report preliminary findings that should be validate in a larger series of patients. At the end, we did not assess the quality of DWI images; this could be considered a future endpoint.

Our results support the hypothesis that Mean of Diffusion Coefficient derived by DKI allows to identify responders and not responder patients subject to ECT treatment. MD had high diagnostic performance to assess ECT response, showing a significant variation of value between pre and post treatment and a significant statistically difference for percentage change between responders and not responders patients.