Introduction

The recommended imaging modality for prostate cancer (PCa) is multiparametric MRI (mpMRI), which includes morphological T2-weighted imaging (T2W), diffusion weighted imaging (DWI) and dynamic contrast-enhanced (DCE) MRI. In the most recent version of the Prostate Imaging Reporting and Data System (PIRADS 2.0), the role of dynamic contrast-enhanced (DCE) has become marginal, and its complete exclusion is currently under debate¹. However, quantitative analysis of the DCE series is currently not performed, overlooking valuable information on cancer angionenesis. Recently, magnetic resonance dispersion imaging (MRDI) was proposed for quantitative spatiotemporal analysis of the DCE-MRI series, showing improved diagnostic performance compared to the standard Tofts model². MRDI is based on modeling both the contrast agent transport in the vasculature and its extravasation in tissue, enabling the simultaneous estimation of parameters related to the microvascular architecture and leakage; additionally, there is no need for the separate measurement of the arterial input function, which is instead required for the application of Tofts’ model, thus facilitating the application of MRDI in clinical practice. In this study, we investigate the potential added value of MRDI in the context of PIRADS by comparing the performance of MRDI with mpMRI for PCa diagnosis.

Methods

This multicenter study included 76 PCa patients from the Prostate Cancer Molecular Medicine database who underwent mpMRI and radical prostatectomy. Institute review board and ethical committee approval was obtained at each institution. For MRDI analysis, time-intensity curves extracted at each pixel were first converted into concentration-time curves by T1 mapping² and then fit by the reduced dispersion model, described as²
$$C(t) = A \sqrt{\frac{\kappa_d}{2 \pi (t- t_0)}}e^{-\frac{\kappa_d(t-t_0-\mu)^2}{(t-t_0)}} \ast e^{-k_{ep}t}$$
where $$$A$$$ is a multiplicative constant, $$$\kappa_d$$$ is the local dispersion parameter, characterizing the microvascular architecture, $$$t_0$$$ is the theoretical injection time, $$$\mu$$$ is the mean transit time, $$$k_{ep}$$$ is the back-flux rate constant, and $$$\ast$$$ represents the convolution operator. Since it is the MRDI parameter with the best classification performance², parametric maps of $$$\kappa_d$$$ were obtained and used in this study. To compare the diagnostic performance of MRDI with standard mpMRI, the prostate was divided into sectors, as described in PIRADS 2.0. Two radiologists (R1, R2) performed three randomized scorings of each sector by evaluating: mpMRI, according to PIRADS 2.0; MRDI parametric maps, according to custom guidelines; mpMRI and MRDI maps together (mpMRI+MRDI). In parallel, the histological ground truth was scored in consensus by 2 pathologists with a custom Likert-based system, based on Gleason score and the surface percentage of cancer in each sector. All scoring systems ranged between 1 and 5. For all scorings, clinically significant PCa (csPCa) was defined per sector as a score>=4. The diagnostic performance for csPCa on a patient-level was assessed by assigning a positive if at least 1 sector was scored >=4, using the pathology scoring as the ground truth.

Results

According to the pathology outcome, csPCa was found in 51 patients. Table 1 provides the diagnostic performance in terms of sensitivity, defined as true positives (TP)/total positives (P), and specificity, defined as true negatives (TN)/total negatives (N). While most csPCa is found with both mpMRI and MRDI (mpMRI ∩ MRDI), adding the detections by moth methods (mpMRI ∪ MRDI) improves the result as some csPCa are only found either with mpMRI or MRDI (Table 2). Similarly, some true negatives (non-significant PCa, benign lesions or no lesion) were correctly recognized only with either mpMRI or MRDI only (Table 3). For radiologist R1, the mpMRI+MRDI scoring was identical to the mpMRI scoring, suggesting that the additional information provided by the quantitative MRDI maps was neglected.


Table 1 Diagnostic Performance

	Sensitivity (TP/P)			Specificity (TN/N)
Radiologist	mpMRI	MRDI	mpMRI+MRDI	mpMRI	MRDI	mpMRI+MRDI
R1	0.78 (40/51)	0.71 (36/51)	0.78 (40/51)	0.36 (9/25)	0.36 (9/25)	0.36 (9/25)
R2	0.59 (30/51)	0.78 (40/51)	0.65 (33/51)	0.68 (17/25)	0.24 (6/25)	0.2 (5/25)

Table 2 Comparison scorings mpMRI and MRDI in terms of Sensitivity

	Sensitivity (TP/P)
Radiologist	mpMRI ∩ MRDI	mpMRI ∪ MRDI	mpMRI only	MRDI only
R1	0.57 (29/51)	0.92 (47/51)	0.21 (11/51)	0.14 (7/51)
R2	0.51 (26/51)	0.86 (44/51)	0.07 (4/51)	0.27 (14/51)

Table 3 Comparison scorings mpMRI and MRDI in terms of Specificity

	Specificity (TN/N)
Radiologist	mpMRI ∩ MRDI	mpMRI ∪ MRDI	mpMRI only	MRDI only
R1	0.16 (4/25)	0.56 (14/25)	0.2 (5/25)	0.2 (5/25)
R2	0.24 (6/25)	0.68 (17/25)	0.44 (11/25)	0 (0/25)

Discussion and conclusions

Although the performance was comparable using mpMRI or MRDI, no improvement was observed when using both mpMRI and MRDI. Yet, in up to 14/51 of patients, csPCa was correctly found only with MRDI, which could potentially lead to a 27% increase in sensitivity (Table 2); additionally, in up to 5/25 patients, non-significant prostate cancer, benign lesions or no lesion were correctly recognized only with MRDI, potentially leading to a 20% increase in specificity (Table 3). This suggests that used in conjunction with mpMRI, MRDI quantitative analysis can improve csPCa diagnosis in the context of PIRDAS. However, scoring mpMRI and MRDI together generally did not improve the performance, highlighting that appropriate radiological training is necessary to learn how to interpret the combined information of mpMRI and MRDI.

Acknowledgements

We acknowledge Chris Bangma for making the Prostate Cancer Molecular Medicine dataset available for this study