Multiparametric magnetic resonance imaging (mpMRI) of the prostate combines T2-weighted
imaging, diffusion-weighted imaging and dynamic contrast-enhanced imaging. Correlation with
radical prostatectomy specimens has demonstrated that mpMRI has excellent sensitivity in
detecting prostate cancers (PCa) with a Gleason score ≥7 and cancers with a Gleason 6 and a
volume ≥0.5 cc. Nevertheless, its specificity is poor and there is large overlapping between
the appearances of benign and malignant prostate lesions. As a result, the use of a 5-point
subjective score has been widely encouraged to describe the level of suspicion of prostate
lesions. This so-called 'Likert score' is a highly significant predictor of the malignant
nature of prostate focal lesions. However, because there are no descriptions of specific
criteria to be used in the scoring process, the Likert score relies heavily on the reader's
experience.
In an attempt to standardize mpMRI interpretation, the European Society of Urogenital
Radiology and the American College of Radiology recently endorsed the so-called Prostate
Imaging-Reporting and Data System (PIRADS) score. The second version of this scoring system
(PI-RADS v2 score) gave good results in characterizing prostate focal lesions. However,
Inter-reader agreement remains moderate at best, even after training, and there is still a
high-rate of false positives. These results have led some authors to suggest that there might
be structural limits to the ability of any score based on MR imaging to allow detection of
prostate cancer with high specificity.
Using quantitative magnetic resonance (MR) image features to characterize prostate lesions
seen on mpMRI could improve interpretation standardization, and recently, several
computer-aided diagnosis (CAD) systems combining various image features have shown promising
results in characterizing prostate tissues. However, most CAD systems have been trained and
evaluated on images from the same MR scanner. Unfortunately, quantification in MR imaging is
limited by substantial inter-manufacturer variability in the calculation of quantitative
image parameters. The quantitative thresholds defined for one manufacturer may therefore not
be valid for another manufacturer. Of the many reported CAD systems, only few have shown
robust results at cross-validation in datasets from different manufacturers.
We developed in Lyon a mpMRI CAD system for discriminating Gleason ≥7 cancers in the
peripheral zone (PZ). That CAD system was trained using mpMRI from patients treated by
radical prostatectomy. It combines the 10th percentile of the apparent diffusion coefficient
(ADC_10th) and the time to the peak of enhancement (TTP) at dynamic contrast-enhanced (DCE)
imaging. It provided good results when cross-validated in two datasets from two different
manufacturers (General Electric and Philips). We then tested the CAD on a cohort of 130
patients who underwent mpMRI (General Electric or Philips MR unit) before prostate biopsy.
Each MR lesion targeted at biopsy had prospectively received a Likert score of likelihood of
malignancy at the time of the biopsy. Retrospective analysis of these MR lesions with the CAD
showed that the stand-alone CAD outperformed the Likert score in predicting the presence of
Gleason ≥7 cancer at biopsy (Area under the receiver operating characteristic curve (AUC):
0.94 (95% confidence interval (95CI): 0.90-0.98 versus 0.81 (95CI: 0.75-0.88), p<0.0002)).
These good results encourage us to perform an external validation of the CAD testing its
performance on mpMRI from another manufacturer (Siemens) and another institution.
The principal objective of the DIJON-CAD study is to evaluate the performances of the QCAD
developed in Lyon (QCAD/Lyon) in a cohort of consecutive patients treated by prostatectomy
and who underwent preoperative mpMRI on a Siemens 3 Tesla MR imager at the Dijon University
Hospital center or at the Dijon Cancer Center (both institutions share the same MR unit).
This study is the first step of the external validation of the QCAD/Lyon system. It is only
aimed at verifying that the diagnostic performance of the system is not very poor on external
mpMRI (which is a substantial risk). If the results are good, a proper multicentric
prospective validation study will be planned.