Purpose

The purpose of this study was to explore the prediction capability of whole-tumor histogram analysis in diffusion kurtosis imaging (DKI) to assess tumor response to neoadjuvant chemoradiation therapy in locally advanced rectal cancer.

Materials and methods

Between June 2016 and November 2017, totally 43 patients with rectal cancer were retrospectively collected in our study. All patients received neoadjuvant chemoradiation (NCRT) followed by total mesorectal excision. All enrolled patients underwent MRI including DKI sequence at two timepoints: 1-7 days before NCRT (pre-DKI), within 1-7 days after NCRT (post-DKI). Pretreatment T stage, N stage, mesorectal fascia (MRF),extramural vascular invasion (EMVI), tumor location and length, pretreatment CEA and CA19-9 levels were recorded. According to tumor response, we classified the patients into pCR and non-pCR groups.The parameters of DKI sequence were as following: b values = 0, 700, 1,400 and 2,100 s/mm²; FOV = 240×180 mm²; scan matrix = 120×120; TE/TR = 4,800/79 ms; bandwidth =1,894 Hz/pixel; ADC map were inline calculated with all acquired b values. DKI derived parameter D and K maps were calculated with all acquired b values by using the prototype Body Diffusion toolbox (Siemens Healthcare, Erlangen, Germany). Whole tumor histograms analysis was performed by using a prototype MR Multiparametric Analysis software (Siemens Healthcare, Erlangen, Germany) on both pre- and post-DKI derived parameter maps. The histogram features including Volume, Mean, Standard deviation (SD), Median(Med), Percentiles (75th, 95th), Skewness(ske), Excess Kurtosis(ExcKur) and Difference Entropy (DiffEnt) were calculated for ADC, K and D maps.Histogram features of the difference between pre-DKI and post-DKI maps were also generated. The relationship between these histogram features, tumor characteristics and tumor response were evaluated by using Logistics regression.

Results

Pretreatment K_Ske, Pretreatment tumor volume, and ADC_SD (pre-post) were correlated with tumor response by using Lasso regression model (Figure 1) and radiological signature score (Rad-score) was built (Figure 2 and 3) as Rad-score = 0.49×Pre-K_ske+0.08×post-tumor volume-0.34×ADC_SD(pre-post)+1.61. Pretreatment EMVI(+) (OR=9.231 (1.015-83.938), p=0.048), MRF(+) (OR=5.317 (1.136-12.488), p=0.030),higher rectal cancer (OR=4.110 (1.234-13.688)，p=0.021), CEA ≥5ng/ml (OR=9.231(11.015-83.938), p=0.048) and low Rad-score (OR=14.428 (1.833-113.588), p=0.011) could be associated with non-pCR in univariate analysis. Multivariate analysis identified that the pretreatment CEA≥5ng/ml (OR=42.311 (1.098-1630.20), p=0.044) and low Rad-score (OR=1.312 (1.077-1.597), p=0.007) would be independent factors to indicate patient with non-pCR.

Conclusion

Whole-tumor histogram analysis of DKI derived parameter maps has potential value to predict tumor response. It will be helpful to stratify patients who are sensitive to neoadjuvant chemoradiation in locally advanced rectal cancer.

Acknowledgements

No acknowledgement found.

References

No reference found.