Synopsis

A modified version of MR dispersion imaging, mMRDI, is introduced to achieve an effective estimation of intravascular dispersion parameter while maintaining low computational complexity. A total of 53 patients who underwent 3T mp-MRI exams prior to robotic assisted radical prostatectomy are included for evaluation. The estimation of the additional intravascular dispersion related parameter offers minimum model fitting errors and improved delineation between clinically significant prostate cancer and normal tissue assessed by the receiver operating curve analysis (AUC = 0.92).

Purpose

Theory

The intravascular dispersion in the estimated AIF was previously described as a simple model in DSC-MRI (1) as a convolution with a vascular transport function, h(t), $$C_p^{mMRDI}(t) = C_p(t)\star h(t)$$ A well-known model for h(t) is to use an exponential decay based on the assumption that the vascular bed is a single, well-mixed compartment (2): $$ h(t) = \frac{1}{\beta}\exp^{\frac{-t}{\beta}},$$ where $$$\beta$$$ [s] corresponds to the effective dispersion (i.e., the larger the $$$\beta$$$, the larger the dispersion). The AIF of mMRDI can be then expressed as the population-averaged AIF convolved with h(t), where h(t) is only controlled by the dispersion factor $$$\beta$$$. Fig 1 demonstrates the AIFs of mMRDI with various $$$\beta$$$ when Parker AIF (3) was used. The intravascular dispersion model, $$$C_p^{mMRDI}(t)$$$, can be combined with the standard Tofts model (4) where the dispersion factor $$$\beta$$$ is part of the free parameter to estimate, $$C_t(t) = K^{trans}\int_{t_0}^{t+t_0}C_p^{mMRDI}(\tau)\exp^{-k_{ep}(t-(\tau-t_0))}d\tau.$$ The relative ratio of the intravascular dispersion, $$$\beta^*$$$, can be also computed by, $$\beta^* = \frac{\beta_{WP}}{\beta},$$ where $$$\beta_{WP}$$$ is the dispersion factor computed from the whole prostate measured by each subject.

Methods

With an IRB approval, a total of 53 patients who underwent 3T mp-MRI exams prior to robotic assisted radical prostatectomy were included for evaluation. 49 clinically significant prostate cancer lesions (Gleason score (GS) $$$\geq$$$ 7) were identified on mp-MRI from genitourinary radiologists and were later confirmed by the corresponding whole-mount histopathology (WMHP). Each corresponding lesion on WMHP was referenced to draw the region of interest (ROI) containing the lesion and classified each MRI lesion as clinically significant and non-clinically significant prostate cancer. Normal tissues in different prostatic zones (TZ: transition zone and PZ: peripheral zone) were also manually defined in order to assess how well the modeled parameters delineate between cancerous and normal tissue.

Three DCE-MRI models were compared, including standard Tofts with two population-averaged AIFs, Weinmann (5) and Parker (3), and mMRDI. The adequacy of three models were first evaluated by the corrected Akaike information criterion (cAIC) (6, 7). A median value within each ROI was calculated after generating pixel-by-pixel parametric and cAIC maps derived from four DCE-MRI analysis models. The linear discriminant analysis (LDA) was used to detect clinically significant prostate cancer from normal tissue in TZ (n=10) or PZ (n=39). The performance of each DCE-MRI analysis model was evaluated by the 10-fold cross validation by averaging of the ten accuracies. The receiver operating curve (ROC) and areas under the ROC curve (AUC) were performed to compare its performance in TZ or PZ.

Results

Fig 2 shows the model comparison by model fitting errors in different tissue types. The boxplots of cAIC show that mMRDI has the minimum median, and first and third quartiles in all tissue types (prostate lesion, TZ, and PZ), suggesting the minimum fitting errors. Fig 3 contains an example of pixel-by-pixel parametric maps using three DCE-MRI analysis models with subsequent mapping of all prostate cancer lesions at WMHP. Two lesions were identified from histopathologic examinations, such as GS 4+5 in the right posterior to anterior and GS 4+4 in the left anterior to posterior. The parametric map from mMRDI shows increased specificity in the delineation of prostate cancer, probably due to an additional estimation of the intravascular dispersion parameter.

Fig 4 shows the ROC analyses in detecting clinically significant prostate cancer in TZ or PZ. The model parameters derived from mMRDI show superior performance, and the delineation between normal and clinically significant prostate cancer tissue was improved in both transition (AUC=0.92) and peripheral zones (AUC=0.92), in comparison with other DCE-MRI analysis models.

Conclusion

Acknowledgements

References

1. Calamante F, Gadian DG, Connelly A: Delay and dispersion effects in dynamic susceptibility contrast MRI: simulations using singular value decomposition. Magn Reson Med 2000; 44:466–73.

2. Lassen NA, Henriksen O, Sejrsen P: Indicator Methods for Measurement of Organ and Tissue Blood Flow. In Compr Physiol. Hoboken, NJ, USA: John Wiley & Sons, Inc.; 2011.

3. Parker GJM, Roberts C, Macdonald A, et al.: Experimentally-derived functional form for a population-averaged high-temporal-resolution arterial input function for dynamic contrast-enhanced MRI. Magn Reson Med 2006; 56:993–1000.

4. Tofts PS: Modeling tracer kinetics in dynamic Gd-DTPA MR imaging. J Magn Reson Imaging ; 7:91–101.

5. Weinmann HJ, Laniado M, Mützel W: Pharmacokinetics of GdDTPA/dimeglumine after intravenous injection into healthy volunteers. Physiol Chem Phys Med NMR 1984; 16:167–72.

6. Brix G, Zwick S, Kiessling F, Griebel J: Pharmacokinetic analysis of tissue microcirculation using nested models: Multimodel inference and parameter identifiability. Med Phys 2009; 36:2923–2933.

7. Wagenmakers E-J, Farrell S: AIC model selection using Akaike weights. .