Synopsis

Keywords: MSK, Cancer, Intravoxel Incoherent motion, Chemotherapy response prediction, Performance comparison

Predictive performance of penalty-based IVIM analysis methodologies BE+TV and BE+HPF for response to Neoadjuvant Chemotherapy have been evaluated using clinical dataset with osteosarcoma in comparison with existing IVIM analysis methods. IVIM datasets before and after chemotherapy were analyzed using 5 IVIM analysis methods – Bi-exponential model and two of its segmented variants and penalty-based BE+TV and BE+HPF methods. Results showed, IVIM parameters estimated by BE+TV and BE+HPF methods produced improved prediction of response to chemotherapy in osteosarcoma than the existing IVIM analysis methods at both the time-points, before and after chemotherapy.

Introduction

Penalty-based intravoxel incoherent motion (IVIM) analysis methods; bi-exponential model with adaptive Total Variation(TV) penalty function (BE+TV) and bi-exponential model with adaptive Huber penalty(HPF) function (BE+HPF); have shown improved IVIM parameters estimation in comparison to the conventional bi-exponential model¹ and its variants using simulation datasets and empirical clinical datasets of bone tumor². In this study, performance of BE+TV and BE+HPF methodologies for predicting response to neoadjuvant chemotherapy (NACT) has been compared with existing conventional IVIM analysis methodologies using clinical dataset of osteosarcoma.

Materials and Methods

Dataset: IVIM MRI datasets of 35 patients (N=35,Age=17.6±2.4 years,M:F=25:10) with osteosarcoma were analyzed. All patients underwent 3 cycles of NACT consisting of Cisplatin and Doxorubicin at every 3 weeks³ followed by surgery. IVIM datasets were acquired at 1) Baseline and 2) after completion of chemotherapy(follow-up) using a 1.5 T Philips Achieva® scanner. IVIM MRI was acquired using free breathing Spin Echo-Echo Planar imaging at 11 b-values (0,10,20,30,40,50,80,100,200,400,800 s/mm²), with TR/TE=7541/67 msec, matrix-size=192×192 and slice thickness/Gap=5 mm/0.5 mm, field-of-view=250×250 mm². Region of interest (ROI) for tumor volume was demarcated an expert radiologist (>12 years of experience in cancer imaging) thoroughly across the slices of DWI with b=800 covering the whole tumor.
Gold standard for NACT response: Histological necrosis was evaluated in resected tumor. ≥50 necrosis was considered as good-responder and <50 necrosis was considered as poor-responder to chemotherapy according to Picci et al⁴.
Quantitative IVIM analysis: Quantitative IVIM parameters Diffusion coefficient (D), Perfusion coefficient (D*) and Perfusion fraction (f) were estimated in tumor volume at baseline and follow-up. Five different IVIM methodologies were used for parameter estimation as: exiting IVIM analysis methods a) Bi-exponential (BE) model, b) Segmented BE method with two-parameter fitting (BEseg-2), c) Segmented BE method with one-parameter fitting (BEseg-1) and penalty based IVIM methods d) BE+TV and e) BE+HPF. Apparent diffusion coefficient was estimated in tumor volume for completion.
Statistical analysis: Average values of quantitative parameters ADC,D,D*,f were evaluated in responder and non-responder groups at baseline and follow-up and compared using student-t test for statistical significance. Inter-scan comparison (between baseline and follow-up) of absolute mean of the quantitative parameters (ADC,D,D*,f) in tumor volume was performed using paired t-test. Receiver operating characteristics(ROC) curve analysis was used to evaluate the predictive performance of all five IVIM methods for NACT response in osteosarcoma at baseline and follow-up. Reproducibility of IVIM methodologies was evaluated using coefficient of variation(CV). A p-value<0.05 was considered as statistically significant.

Results

Table1 present the average parameter values in tumor volume evaluated using these five IVIM methodologies among responder and non-responder groups at baseline and follow-up. In responder group average D in tumor were comparatively lower than the non-responders (1.06-1.24×10^-3mm²/s vs. 1.22-1.41×10^-3mm²/s) at baseline and significantly increased after NACT (1.47-1.72×10^-3mm²/s & 1.52-1.73×10^-3mm²/s respectively). D values between responder and non-responder groups were not significantly different (p>0.05) at baseline and follow-up.
At baseline, average D* in tumor among non-responders were comparatively higher than responders (30.95-42.24×10^-3mm2/s vs. 23.76-39.21×10^-3mm²/s) and significantly reduced after NACT (23.44-36.44×10^-3mm²/s & 17.77-31.48×10^-3mm²/s respectively). D* values between responder and non-responder groups were not significantly different (p>0.5) for all other methods except BE+TV and BE+HPF at baseline (23.76±7.56×10^-3mm²/s vs 30.95±10.80×10^-3mm²/s;p=0.04).
In responder and non-responder groups average f in tumor at baseline were in the range 11.43%-30.9% and 11.92%-19.36% respectively and did not change significantly after NACT (8.76%-33.01% and 8.75%-27.21% respectively). Average f values between responder and non-responder groups were not significantly different (p>0.5) for all other methods except BE+TV and BE+HPF at follow-up (13.2±1.3% vs 12.0±0.9%;p=0.03).
Table2 shows the CV values for IVIM parameters for 5 analysis methods at baseline and follow-up. Figure1 and Figure2 depict the IVIM parametric maps evaluated using 5 analysis methods at baseline and follow-up for a representative patient each from responder and non-responder group.
ROC curve analysis for NACT response prediction using 5 analysis methods has been summarized in Table3 and Table4 at baseline and follow-up respectively. At baseline, D* evaluated using BE+TV and BE+HPF methods showed highest AUC=0.7 with sensitivity=65%, specificity=75% at a threshold value of >26.9×10^-3mm²/s. At follow-up, f values evaluated using BE+TV and BE+HPF methods showed highest AUC=0.76,0.77 respectively with sensitivity=77%, specificity=69% at a threshold value of >15.30%. D,D*,f evaluated using BE+TV and BE+HPF method combinedly produced highest AUC=0.75,0.8 with sensitivity=72%,77%, specificity=77%,75% for predicting NACT response at baseline and follow-up respectively. Figure3 demonstrates the ROC curve analysis for predictive performance of 5 analysis methods for NACT response at baseline and follow-up.

Discussion

Estimated CV for BE+TV/BE+HPF methods were comparatively lower than BE, BESeg-2 and BESeg-1 methods and were comparable with the other spatial constraint methods^5,6. Because of the adaptive nature of penalty functions TV/HPF, the parametric images evaluated by the BE+TV and BE+HPF methods are comparatively less noisy and observed to have higher reproducibility for D* and f parameters as well as improved predictive performance for NACT response than the existing IVIM analysis methods. There was no significant statistical difference between BE+TV and BE+HPF methods. The results were evaluated on limited number of clinical datasets; however, experiment on larger datasets is desirable.

Conclusion

Penalty-based IVIM analysis methods BE+TV and BE+HPF demonstrated improved predictive performance for NACT response using osteosarcoma datasets in comparison to the existing BE, BESeg-2 and BESeg-1 methods.

Acknowledgements

No acknowledgement found.