The usefulness of MRI for imaging prostate carcinoma has already been recognized [7]. On the other hand, most radiologists would agree that the diagnosis and localization of carcinoma are not always easy due to coexisting hyperplasia, prostatitis, or bleeding, and it is sometimes difficult to fill in medical records in a “black or white” manner. Recently, two scoring systems were recommended: PI-RADS and the Likert scale. The PI-RADS system uses multi-parametric techniques including T2-weighted imaging, DCE MRI, and DWI, and a score from one to five is given according to each variable [1]. Therefore, total scores range from 3 to 15, and a threshold of 8 or greater, or 9 or greater, has been used as a cutoff for cancer detection in previous studies [8, 9]. Regarding the Likert system, a rating from 1 to 5 was assigned based on the overall impression of MRI findings, and a threshold of 3 or higher was typically used in previous studies [3–5]. In the present study, we performed a close comparison between all small cancerous foci on whole-mount histopathology and radiological findings according to 5-point Likert scaling using the recent 3.0-Tesla multiparametric MRI. We observed that the proportion of >0.5 cm3 tumors increased according to the upgrade of Likert scores (score 1 or 2: 33 %; score 3: 68.8 %; score 4 or 5: 90.9 %, χ2 test, p < 0.0001), and the proportion of those with Gleason score >7 also increased from score 2 to score 5 (score 2: 0 %; score 3: 56.3 %; score 4: 72.7 %; 5: 90.9 %, χ2 test, p = 0.0001). Our observations confirmed that a threshold of 3 or higher is very helpful for clinicians when considering the possibility of significant cancer, denoting >0.5 cm3 or Gleason score >7 tumors. Although we did not assess PI-RADS data or inter-observer variability in scoring, Renard-Penna et al. reported favorable interobserver agreement between the Likert scale (κ = 0.80) and PI-RADS system (κ = 0.73) [8].
Regarding detectability on MRI according to the cancer volume, Ikonen S et al. previously reported that, with the use of endorectal coil 1.5-T MRI (T2-weighted), the rate of detecting carcinoma foci smaller than 5 mm was 5 %, but it was 89 % for those larger than 10 mm [10]. Roethke MC et al. reported similar results, whereby they were able to visualize 0/56 lesions with a size of <0.3 cm (0 %), 4/116 (3 %) between 0.3 and 0.5 cm, 22/169 (13 %) between 0.5 and 1 cm, 61/136 (45 %) between 1 and 2 cm, and 50/56 (89 %) >2 cm using endorectal coil 1.5-T MRI (T2-weighted) [11]. Villers et al. also reported that sensitivity, specificity, and positive and negative predictive values for cancer detection by 1.5-T pelvic phased-array coil MRI were 90, 88, 77, and 95 %, respectively, for foci larger than 0.5 cc [12]. In the present study, the positive predictive value for a diagnosis of cancer based on MRI findings was 75 % (15/20) for score 3, 73.3 % (11/15) for score 4, and 95.7 % (22/23) for score 5. Using a threshold of 3 or greater to indicate probable cancer, the detection rate on MRI markedly improved (0.5–1.0 cm3: 6/9, 66.7 %, 1.0 < cm3: 35/38, 92.1 %) when the tumor foci volume exceeded 0.5 cm3. Overall, sensitivity for cancer detection was 87.2 % (41/47) for tumors larger than 0.5 cm3, and the positive predictive value was 82.8 % (48/58) for score >3. Multivariate analysis identified only the tumor volume as being significantly correlated with visibility on MRI. Because 1 cm in diameter represents a sphere of 0.5 cm3, we consider that, although the detection limit of small foci was the same as that using endorectal coil or pelvic phased-array coil 1.5-T MRI, modern 3.0-Tesla multiparametric MRI offers a more sophisticated image of the prostate and can clearly visualize most tumors larger than 0.5 cm3. In other words, 3.0-Tesla can identify clinically significant disease in terms of a tumor volume >0.5 cm3. In contrast, it can barely detect tumors with a volume of less than 0.5 cm3 regardless of the Gleason score. In the current study, an endorectal coil was not used, and we, therefore, did not generate data on how an endorectal coil can aid in tumor depiction. Previously, Park BK et al. and Sosna J et al. reported that 3.0-T pelvic phased-array MRI could produce an image equivalent to 1.5-T endorectal MRI [13, 14]. Kim BS et al. also reported that the staging ability was not significantly different between 3.0-T pelvic phased-array MRI and 3.0-T endorectal coil MRI [15]. In contrast, Turkbey et al. recently reported that the combined use of a nonendorectal coil and an endorectal coil led to the detection of more cancerous foci than the sole use of a nonendorectal coil [16]. At present, we consider that patients’ discomfort and the additional cost are drawbacks to the use of an endorectal coil, and preoperative evaluation with a nonendorectal coil would be more acceptable universally in daily clinical practice.
As described above, the positive predictive value for a diagnosis of cancer based on MRI findings was 75 % (15/20) for score 3, 73.3 % (11/15) for score 4, and 95.7 % (22/23) for score 5. Conversely, 10 lesions assigned a score >3 (3: n = 5; 4: n = 4; 5: n = 1) were false-positive findings on MRI. After the analysis, we convened a meeting with the pathologists to discuss the cause of these false-positive findings, and found that the most frequent histological finding was inflammation (n = 7), and abscess formation was also noted in one case (data not shown). It would be optimal to perform MRI before prostate biopsy to minimize these artifacts. It has been established that an interval of at least eight weeks is needed to minimize artifacts due to prostate biopsy [17]. Regarding the 6 foci greater than 0.5 cm3 which were missed on MRI (Fig. 3), a strong background due to hemorrhage (n = 1), an anterior TZ location (n = 1), an apex location (n = 1), and islet distribution of small foci (n = 3) may have compromised visualization on MRI. Regarding the relationship between the tumor location and MRI visibility, previous studies showed the lack of additional benefit of DCE-MRI and the difficulty of cancer detection in the transition zone [18, 19]. In the present study, as shown in Table 2, although the tumor location was not associated with visibility on 3-T MRI, our sample size was too small to draw a definitive conclusion, and we agree that the co-existing prostatic hyperplasia might compromise cancer detection in the transition zone.
Our study has several potential limitations. Firstly, this was a small retrospective study. There was a selection bias in that only patients undergoing prostatectomy were enrolled. As described above, because all MR images were reviewed in consensus by two radiologists during a single session, we do not have data on inter- or intraobserver variability of the Likert scale. The lack of an endorectal coil may have influenced our observations. Furthermore, in cases undergoing MRI evaluation after biopsy, post-biopsy hemorrhage might have resulted in difficulty of tumor detection. It remains unknown whether the results would be reproducible in a different institute with a reader with a different experience level. Further validation studies are warranted to evaluate whether a scoring system including a Likert scale can become a common language among physicians treating prostate carcinoma patients in daily clinical practice.