False positive and false negative diagnoses of prostate cancer at multi-parametric prostate MRI in active surveillance

A detailed understanding of the normal zonal anatomy of the prostate is essential for interpretation of prostate MRI. In 1981, McNeal [23] described the three distinct prostate zonal regions: (1) the peripheral zone, (2) transition zone and (3) central zone (CZ). The prostatic zonal anatomy is best depicted at T2-weighted (W) MRI (Fig. 1). The PZ is hyperintense on T2W because of abundant glandular tissue, is located at the periphery of the gland and harbours 70 % of PCa [24] (Fig. 1). Previously, it was thought that the TZ [the site of benign prostatic hyperplasia (BPH)] and the CZ (which surrounds the ejaculatory ducts, is located mainly at the base of the prostate, is posterior to the TZ and the urethra, and is proximal to the verumontanum) could not be differentiated at imaging and these areas were collectively referred to as the central gland [25, 26]. The central gland in most adult males consists of hypertrophied TZ that compresses the CZ against the surgical capsule [25]. Currently, it is acknowledged that the CZ can be identified separately from the TZ in up to 4/5 of males and the CZ appears as a symmetric band of homogeneously low signal intensity (SI) on T2W MRI and apparent diffusion coefficient (ADC) maps best seen at the prostate base [25], (Fig. 1).

Accurate identification of the CZ is critical as, in our experience, many false-positive interpretations of PCa occur when mistaking the CZ for PCa. The CZ can be diagnosed by noting its symmetry (particularly on coronal T2W images) and typical location adjacent to the ejaculatory ducts [25, 26]. Functional imaging techniques also aid in the differentiation of CZ from PCa. In the normal CZ, low SI on ADC is due to inherently low T2 SI not true restricted diffusion (“T2 black hole effect”); see Technical Pitfall 2. In the normal CZ, there is no true diffusion restriction; therefore, there should be no increased SI on long (b ≥1000 mm²/s) EPI (Figs. 1 and 2). The enhancement pattern of the CZ at DCE has not, to our knowledge, been described; however, in our experience, a progressive type 1 enhancement is typical of the CZ (Fig. 1). CZ PCas are aggressive [26‐28], but account for less than 5 % of PCa [26]. An asymmetry of the central zone should raise suspicion for a possible tumour [25]; however, it is more common to observe a slight asymmetry in the thickness of the CZ because of the orientation of the prostate gland and the plane of imaging (Fig. 2). To diagnose the rare CZ PCa, asymmetry at T2W should be confirmed in multiple planes and corresponding functional imaging findings such as increased signal intensity on long b value trace EPIs and/or suspicious contrast kinetics at DCE are helpful, as these latter findings are not expected in the normal CZ (Fig. 2). Ultimately targeted biopsy or MRI follow-up may be necessary for suspicious areas in the CZ.

The anterior fibromuscular stroma (AFMS) is a normal structure at the extreme anterior midline of the prostate gland, which is inseparable from the surrounding normal prostatic stroma [25, 29]. The AFMS contains no glandular tissue and is composed of widely spaced smooth muscle cells. It appears homogeneously hypointense on T2W MRI and because of this inherent low T2W SI will appear as low SI on ADC maps. Differentiation of the AFMS from anterior cancer is established by noting the midline location, well-defined margins, lack of true diffusion restriction (absence of high SI on b ≥1000 mm²/s EPI) and a benign progressive type 1 enhancement pattern at DCE (Fig. 3) [25, 30]. PCa arising in the extreme anterior prostate can resemble the AFMS and in many cases anterior tumors can extend up to or invade the AFMS. Both lesions are low SI on T2W; however, anterior PCa is more lenticular or polygonal in shape, has ill-defined or smudged borders, and the bulk of the lesion tends to be off midline (Fig. 4) [31]. The use of functional imaging sequences is contributory (Fig. 4) [30]. DWI increases accuracy for detection of anterior cancers because anterior tumours will be of low SI on ADC because of true restricted diffusion, demonstrating increased SI on trace b ≥1000 mm²/s EPI [32]. The utility of DCE in the TZ and anterior prostate is controversial [33]; however, authors have demonstrated more aggressive enhancement patterns in anterior PCa [34‐37]. In our experience, DCE can be beneficial to discriminate anterior PCa from AFMS; however, overlap with stromal BPH limits its utility in the TZ (discussed later) [35].

Detection of PCa on T2W is predicated on the ability to discriminate low T2 SI tumour from the normal adjacent high T2 SI PZ [15, 38‐40]. The glandular tissue of the PZ contains high concentrations of citrate, which is a natural anti-coagulant; therefore, post-procedural haemorrhage is commonly observed at MRI after prostate biopsies and can persist for up to 4 months after the procedure [41]. Areas of post-biopsy haemorrhage are characteristically hypointense on T2W MRI in up to 80 % of cases and can mimic PCa [41] (Fig. 5). This pitfall is easily avoided by cross-referencing T2W to pre-contrast T1W imaging because areas of haemorrhage will also be characteristically hyperintense on T1W (Fig. 5).

Many institutions continue to defer prostate MRI studies until 3–8 weeks have passed after the biopsy to reduce the amount of post-biopsy haemorrhage and facilitate interpretation. With the advent of functional imaging sequences post-biopsy haemorrhage is, in our experience, much less problematic. Areas of tumour have been shown to be reliably differentiated from haemorrhage using DWI, and DCE analysis can be interpreted through image subtraction [42] (Fig. 5). Moreover, because PCa contains less citrate than the normal PZ, areas that are devoid of haemorrhage in the PZ after biopsy may actually highlight the tumour on T1W imaging (“haemorrhage exclusion sign”) [42]. This imaging finding has recently been shown to have >95 % positive predictive value for tumour localisation when combined with characteristic imaging findings at T2W MRI [42].

BPH is extremely common in the TZ and its prevalence increases with age. BPH can be nodular and TZ nodules are commonly encountered at MP-MRI [35], and nodular BPH may mimic TZ PCa [25]. Nodular BPH is categorised into three main subtypes: glandular, stromal and mixed.

Glandular BPH is characterised histologically by hyperplasia of glandular tissue with papillary buds, infoldings and cysts. Glandular BPH is readily differentiated from TZ PCa at MP-MRI [35] and typical features include a well-circumscribed, sharply demarcated and rounded shape, a continuous low T2W SI rim, increased T2W SI often with cystic change, low SI on long ≥1000 mm²/s b value EPI, “T2 shine-through” on ADC maps and benign/progressive/type I contrast kinetics with a low transfer constant on DCE (Fig. 6) [25, 35].

The presence of increased smooth muscle, lymphocytes and ducts (not associated with prostatitis) and reduced elastic tissue characterises stromal BPH at histopathology. Stromal BPH nodules can be difficult to differentiate from TZ PCa [35] (Figs. 6 and 7). Stromal BPH is typically of low T2W SI, similar to TZ PCa. Stromal BPH is also of low signal intensity on ADC maps because of both inherently low T2W SI and true restricted diffusion (due to compact cellularity) and may or may not have increased SI on long b ≥1000 mm²/s trace EPI. Furthermore, ADC values in stromal BPH are low and overlap with TZ PCa [25, 35]. DCE is limited to differentiate stromal BPH from TZ PCa, because enhancement patterns of stromal BPH overlap with TZ PCa [33‐35]. Previous authors have reported lower ADC and higher K-trans values in TZ PCa compared to stromal BPH, but substantial overlaps exist (Figs. 6 and 7) [35]. Other differentiating features that are more suggestive of stromal BPH include well-defined sharply demarcated margins, a continuous low T2W SI rim and a rounded shape (Figs. 6 and 7) [31, 33].

At histopathology and MP-MRI, it is more common to identify a TZ BPH nodule with features of both glandular and stromal BPH. Typically one pattern (glandular or stromal) will predominate; however, features of both types of BPH can be seen within the same nodule, which can further complicate the diagnosis. In cases of mixed BPH, a combination of imaging features of glandular and stromal BPH are encountered [31]. Usually, the presence of increased T2W SI and cystic change (from glandular BPH) within a lesion are reassuring findings of BPH (Fig. 7) since these findings are rarely encountered in PCa.

Chesnais et al. previously demonstrated that the majority of TZ cancers involve the anterior and apical third of the gland [43]. At histopathology, Bouye et al. previously demonstrated that anterior tumours tend to grow anteriorly and invade the anterior fibromuscular stroma but rarely extend to the posterior PZ [44]. It should be noted that in some instances, when there is uncertainty at MP-MRI regarding the diagnosis of TZ PCa versus stromal BPH due to overlap in imaging features, a targeted biopsy or follow-up MP-MRI may be suggested (Fig. 6).

Acute bacterial prostatitis (ABP) is commonly encountered in clinical practice and often managed conservatively without imaging or surgical intervention [38]. When patients with ABP are imaged in the acute setting, prostatitis can mimic PCa at MP-MRI, demonstrating low SI on T2WI, restricted diffusion [45] and enhancement patterns that overlap with PCa [25] (Fig. 8). Although PCa has been reported to demonstrate more well-defined borders and nodular appearance compared to prostatitis and lower ADC values [45, 46], aside from the presence of abscess formation in prostatitis, differentiation is mainly based on clinical grounds.

Granulomatous prostatitis (GP) is a benign inflammatory entity that may also be indistinguishable from PCa at MP-MRI, demonstrating low SI on T2W, restricted diffusion and suspicious enhancement at DCE [47]. Moreover, granulomatous prostatitis can involve the periprostatic fat or seminal vesicles, mimicking extraprostatic spread of PCa (Fig. 9). With the correct clinical history [e.g., previous bacille Calmette-Guerin therapy for bladder cancer, tuberculous prostatitis or previous intervention such as transurethral resection of the prostate (TURP)], granulomatous prostatitis may be favoured over PCa [25]; however, histological confirmation is often required [47].

In the chronic setting, areas of prostatitis can mimic PCa because they will demonstrate low T2W SI [48]. Functional imaging techniques can be useful to differentiate areas of chronic prostatitis from cancer (Fig. 10) but there is overlap in imaging features [49, 50] and targeted biopsy or MP-MRI follow-up of the suspicion region may be considered depending on the level of suspicion for tumour [47].

PCa is divided into the most common acinar adenocarcinoma and less common non-acinar subtypes. Ductal adenocarcinoma (DCa) is an aggressive variant of PCa and is the most common of the non-acinar subtypes [51] with a reported incidence of 0.5–6 % [52]. DCa is associated with higher rates of positive surgical margins, extraprostatic extension, vascular invasion, seminal vesicle invasion and metastases [53, 54]. A diagnosis of DCa at needle biopsy requires definitive therapy and is considered an absolute contraindication for AS [55, 56]. Schieda et al. recently demonstrated that DCa is of increased SI on T2W MRI, which can render the tumour occult [39]; however, the same authors demonstrated that DCa resembles Gleason score ≥7 or higher tumour on DWI and DCE [57]. Therefore, in a lesion with aggressive imaging findings on DWI/DCE and paradoxical increased T2W SI, DCa should be considered [57] (Fig. 11).

As discussed, detection of tumour in the PZ on T2W MRI is based on differentiation of low T2W SI tumour from the high T2W glandular tissue [15, 38‐40]. In the TZ, T2W imaging is now recognised as the most important pulse sequence for detection of PCa because of overlap between TZ cancer and stromal BPH on functional imaging (see Interpretive pitfalls, stromal BPH) [58]. T2W imaging is typically performed using turbo/fast spin-echo techniques [39]. Motion correction with radial acquisition (BLADE, Siemens Healthcare, Malvern, PA, USA; PROPELLER, General Electric Healthcare, Milwaukee, WI, USA) sequences are popular for pelvic imaging because they correct for in-plane rotation and translational artefacts and have been shown to result in improved overall image quality when compared to conventional spin echo [59, 60]. A disadvantage of these sequences is decreased image contrast compared to conventional spin echo [59]. Recently, Rosenkrantz et al. demonstrated that a minority of PCa foci may be obscured at T2W MRI when BLADE/PROPELLER is used compared to conventional spin echo [61] (Fig. 12).

DWI improves the detection of cancer foci in the PZ and TZ, and ADC values derived from DWI have been repeatedly shown to inversely correlate with Gleason grade of PCa [18, 38]. DWI is the single best imaging test for analysis of the PZ and is used in conjunction with T2W MRI for evaluation of TZ lesions [58, 62]. The visual analysis of DWI must include assessment of trace EPI and should not rely solely upon review of the derived apparent diffusion coefficient (ADC) map. Structures that are inherently of low T2W SI will appear dark on the ADC map because of an inherently low T2W SI (T2 “black-hole” effect) and not from true restricted diffusion (Figs. 1 and 2). Review of trace EPI is critical to avoid this common pitfall because these structures will not appear bright on long (b ≥1000 mm²/s) EPI (Figs. 1 and 2). Conversely, PCa (and as discussed some stromal BPH nodules) is of low SI on ADC and will appear bright on the long (b ≥1000 mm²/s) EPI (Figs. 3, 6 and 7). The use of longer b values (b ≥1000 mm²/s) facilitates the detection of cancer foci compared to the benign prostatic tissues [63, 64]. Visual analysis of ADC may be further limited by automatic display settings and can result in the failure to detect PCa or an underestimation of tumour grade if not displayed properly (Fig. 13). A previous study demonstrated that by using a window width = 1.650 and level = 1.675 × 10–6 mm²/s for display of ADC, higher grade tumours were more likely to appear dark (Fig. 13) [65].

It has been demonstrated repeatedly that ADC values inversely correlate to PCa grade [18]. The use of ADC values across MR systems is problematic because of differences related to magnet strength, vendor system and selection of differing b values [57]. Furthermore, a variety of methods to quantify ADC have been described including single region of interest measurement versus whole lesion analysis and mean, median and complex histogram analysis of ADC. Application of quantitative ADC thresholds may be limited by these factors. A recent study demonstrated that the use of an ADC ratio (normalised to adjacent normal PZ) provided a better inter-scanner comparison than the absolute ADC value [57]. Mean ADC values may also be limited due to the variable composition of prostate cancers and when comparing intermediate prostate cancers (Gleason score 3 + 4 = 7 vs. 4 + 3 = 7) [66] and studies have demonstrated that the use of the low percentile ADC may be a better reflector of the Gleason score [67, 68]. Regardless of the method used, due to differences across systems, at present, we suggest that the use of quantitative ADC in clinical practice should be applied carefully, derived and validated on an institutional basis.

DCE improves the detection of PCa and DCE parameters correlate with PCa grade [69‐71]. Currently, there is no established interpretation criteria for DCE analysis which varies from simple visual analysis to semi-quantitative analysis to full quantitative pharmacokinetic modelling [72]. A recent study demonstrated that DCE is underutilised in clinical practice compared to DWI and that semi-quantitative and quantitative analyses were not commonly used [72]. In the revised PI-RADS v2.0 guidelines, DCE analysis has been modified from semi-quantitative curve analysis to a simple visual analysis [58].

Moreover, DCE is very limited for the detection and grading of cancers in the transition zone because of significant overlap with stromal BPH (Fig. 7). Analysis of potential tumour foci in the transition zone should preferentially favour T2W and DWI over DCE. It is currently recognised that the utility of DCE in clinical practice is mainly as a confirmatory sequence for PZ lesions [58]; however, further studies are required to determine the value of DCE in PCa.

MP-MRI for PCa has transformed practice, but has created new challenges. Obtaining accurate histological correlation from lesions detected at MP-MRI may be challenging. MRI guided biopsy is accurate; however, this technique is limited by cost, availability of and access to MRI, prolonged procedure times and patient discomfort [8]. The use of TRUS guidance for biopsy of MP-MRI-detected lesions is currently a preferred option.

Using existing technology, so-called cognitive registration (CR) is performed. CR requires the TRUS operator to mentally integrate MP-MRI with real-time TRUS, identify suspicious MR lesions and, with TRUS guidance, target those lesions for biopsy. Published data regarding the applicability of CR are limited to institutional series and are highly subject to operator experience [8]. TRUS may be limited for the detection of tumours in the PZ and is also limited for assessment of the TZ and the anterior gland in many patients with enlarged glands due to poor ultrasound beam penetration. A failure to recognise the limitations of TRUS when performing CR biopsies can result in a failure to adequately sample suspicious areas on MP-MRI and ultimately result in delayed diagnosis and therapy for clinically significant PCa (Figs. 14 and 15). In our experience, cognitively registered TRUS biopsy of MP-MRI-detected lesions can be successful, provided that the TRUS operator recognises the limitations of TRUS for demonstrating a corresponding lesion to that seen on MP-MRI. It is not uncommon for a lesion detected in the PZ or TZ on mp-MRI to be sonographically occult on TRUS and, in these instances, by oversampling the corresponding location using anatomic landmarks, the diagnostic yield can be significantly improved (Fig. 16).

The use of Fusion software (which automatically integrates MP-MRI data with real-time 3D TRUS images) provides an alternative to cognitive registration [8]. Fusion software is not without its own limitations including mainly errors in fusion that relate to the spatial deformation of the prostate at TRUS compared to mp-MRI [8]. Moreover, this technology is expensive and at the moment is available in a few specialised centres, although availability is increasing. Studies comparing cognitively registered targeted TRUS biopsy to software fused targeted TRUS biopsy are lacking and have shown mixed results. While several studies have shown no difference between CR and Fusion software [73, 74], other studies have shown an improvement in the detection rate of cancer using fusion software systems [75, 76].

It is critical to emphasise that when a targeted TRUS-guided biopsy performed for a suspicious lesion detected on MP-MRI (using either cognitive registration or fusion software) is negative, the MP-MRI should be reviewed in the context of the biopsy results and other clinical factors in order to consider the possibility of an erroneous targeted biopsy. In these instances, repeat MRI or targeted biopsies can be contemplated.