Cancer still ranks as the leading cause of death and is an important barrier to increasing life expectancy in every country in the world. The global cancer statistics in 2020 showed an estimate of 19.3 million new cases and 10 million cancer deaths. On third place (both sexes) was prostate cancer (PCa) with 7.3% of all new cases and accounting for 3.8% of all cancer related deaths. Among males, PCa comprises 14.1% of newly diagnosed cases and results in 6.8% of cancer-related deaths. PCa ranks second in terms of occurrence, preceded only by lung cancer in 2020. It stands as the fifth most common cause of cancer-related mortality among men in the same year. The incidence rates are three times higher in transitioned countries compared to transitioning countries, with rates of 37.5 and 11.3 per 100,000, respectively. Conversely, the variability in mortality rates is less pronounced, with rates of 8.1 and 5.9 per 100,000, respectively.1

The primary treatment modalities frequently employed for PCa are radical prostatectomy (RP) or radiotherapy (RT), including external beam RT (EBRT) or brachytherapy. Following these treatments, biochemical recurrences (BCR) of PCa can be observed in a range of 27% to 53% of patients, with the occurrence influenced by factors such as a high Gleason score (GS), a short prostate-specific antigen doubling time (PSAdt), and the status of surgical margins or the presence of locally advanced disease.

Thus, the primary diagnosis and staging of PCa is of utmost importance for treatment planning and prognosis. PCa is typically suspected based on digital rectal examination (DRE) and/or PSA levels. To establish a definitive diagnosis, histopathological confirmation of adenocarcinoma in prostate biopsy cores is required.

Current guidelines recommend using transrectal ultrasound (TRUS) in addition to DRE for PCa detection. Furthermore, new sonographic modalities like micro-Doppler, sonoelastography, and contrast-enhanced US (multiparametric US), along with MRI, are suggested for biopsy planning. For clinical staging, including T-staging, N-staging, and M-staging, the recommended imaging modalities are MRI, CT, bone scan, and PET. However, PET imaging in the context of PCa involves combining choline, fluoride, and PSMA PET scans with CT scans, in addition to using MRI. However, hybrid imaging with PET/MRI is not yet included in the current guidelines for PCa evaluation.2

In the context of PET/MRI imaging for PCa evaluation, there are several notable disadvantages that warrant consideration. Firstly, the PET/MRI approach is associated with higher costs compared to PET/CT, making it a potentially less economical option for most healthcare settings. Another drawback lies in the longer scan time required for PET/MRI examinations. While PET/CT scans typically take approximately 20 minutes, PET/MRI scans are clearly longer, leading to potential challenges in patient compliance and overall throughput in busy clinical settings. Moreover, the scan protocols for PET/MRI are considerably more complex than those used in PET/CT. These intricate protocols demand a higher level of expertise and experience from imaging specialist to ensure accurate and efficient execution of the imaging procedures. Another significant limitation of PET/MRI is the complexity and time-consuming nature of interpreting the acquired data. In particular, the evaluation of whole-body staging with PET/MRI poses substantial challenges, as the datasets generated consist of approximately 300% more images compared to PET/CT datasets. This increased volume of data can lead to prolonged interpretation times and potentially impact the workflow and clinical decision-making process.

In 2016 Spick et al. compared 18F-FDG PET/CT and PET/MRI in more than 2300 cancer patients and described it as equally well. So why PET/MRI?

In comparing PET/MRI to PET/CT, the term “equally well” implies that PET/MRI is generally considered inferior due to factors like cost, complexity, and longer examination times. Therefore, PET/MRI must offer significant advantages over PET/CT to justify its implementation. One of the primary advantages of PET/MRI lies in cases where MRI outperforms CT. The superior soft tissue contrast and multiparametric capabilities of MRI make it more suitable for certain indications. Moreover, there are scenarios where both MRI and PET offer distinct advantages, presenting an opportunity for PET/MRI to combine these strengths and provide a comprehensive assessment. A logistic advantage of PET/MRI is the “one stop shop” approach, where patients can undergo a PET/MR scan and receive consolidated findings in a single session, streamlining the diagnostic process. The true added value of PET/MRI stems from its precise image fusion capabilities, enabling the seamless integration of PET metabolic information with MRI anatomical details. This synergy enhances diagnostic accuracy, aiding in better treatment planning and patient management.

In the following, we will briefly explore the history of PET/MRI, tracer development, and its applications concerning PCa. We will also touch technical aspects of PET/MRI and explore the emerging possibilities of utilizing artificial intelligence (AI) in this context.

In 1997, a first preclinical prototype was proposed simultaneous preclinical PET/MR imaging.3 Since then, the technological and methodological aspects of combining PET and MRI have been refined: initially in small animals and later in humans starting from 2010 with a brain insert.4 This differs from PET/CT, which initially addressed a clinical need before being adapted for small-animal imaging. While PET/CT has been remarkably successful. Within the first 3 years of its market introduction in 2001, more than 500 PET/CT systems were installed. Today, several thousand systems from multiple vendors are available worldwide—but the adoption of PET/MRI has been slow. Following the introduction of the first prototype for brain imaging and the subsequent installation of five more units worldwide, the first commercially available whole-body PET/MRI system was introduced in 2011. Fully integrated PET/MRI systems were introduced in 2011 and 2014. Even after 5 years since its introduction, only around 70 PET/MRI systems have been placed worldwide.5 The adoption of PET/MRI systems has been relatively slow, which can be attributed, in part, to the high investment and service costs involved and the need to demonstrate incremental benefits compared to PET/CT or sequential PET/CT and MRI approaches.

In oncology, 18F-FDG, a radioactively labeled glucose analogue, is the most commonly used radiotracer. However, for PCa imaging, 18F-FDG-PET has limited utility as only aggressive, poorly differentiated, or undifferentiated PCa show increased glucose metabolism. In Europe, radioactively labeled choline derivatives were used in the past as radiotracers for PCa imaging. However, due to their nonspecific uptake in benign prostate conditions, they have mainly been used for restaging after primary therapy and for primary staging of high-risk PCa patients. A meta-analysis by Evangelista et al. reported a high specificity of 95% but only a low sensitivity of 49% for primary N-staging.6,7

To address this challenging issue, in the early 2010s another molecular imaging technique called prostate-specific membrane antigen (PSMA) PET has been introduced into clinical practice.8,9 This PET tracer capitalizes on the highly specific expression of PSMA in PCa cells. PSMA, a transmembrane type II glycoprotein, is significantly overexpressed in PCa cells and exhibits increased levels with higher grades, metastasis development, and disease recurrence. A series of studies has demonstrated the superiority of this new technique compared to conventional imaging methods in primary staging and detecting the location of recurrence.10

The most common used PSMA-tracer is 68Ga PSMA-11 as it was approved but the US Food and Drug Administration (FDA) and European Medicines Agency (EMA). Besides this several other PSMA ligands have been clinically established. These are: 68Ga-PSMA-I&T, 68Ga-PSMA-617, 18F-DCFBC, 18F-CFPyL, 18F-PSMA-1007, 18F-JK-PSMA-7, and 18F-rhPSMA-7. The latter was recently been approved by the FDA.11 Indeed, in recent years, there has been a growing inclination towards substituting 68Ga-labeled ligands with 18F-labeled counterparts. The integration of 18F-PSMA in clinical practice presents a range of compelling advantages. Notably, the extended half-life of 18F (110 minutes) surpasses that of 68Ga (68 minutes), improving flexibility in clinical settings. Furthermore, the cyclotron-based production of 18F-PSMA enables costeffective centralized manufacturing and distribution, amplifying its feasibility and accessibility. A noteworthy feature of 18F-PSMA is its potential to enhance spatial resolution due to the shorter positron range, which contributes to improved imaging precision. Additionally, 18F-PSMA exhibits enhanced tumor uptake, and depending on the specific PSMA ligand employed, it demonstrates reduced or even negligible accumulation in the urinary tract, minimizing potential confounders. Preliminary retrospective studies have yielded promising results, underscoring the improved detection capabilities of 18F-PSMA in identifying small metastases or local recurrences, particularly in proximity to the bladder.7,12

While MRI offers enhanced soft tissue contrast and greater sensitivity in detecting bone metastases in PCa, a combined approach utilizing PSMA PET and multiparametric MRI (mpMRI) allows for the simultaneous or sequential acquisition of PET and MR data in a single examination. The mpMRI typically consists of four MRI sequences, namely T2-weighted images, T1-weigthed images, diffusion weighted images (ADC) and dynamic contrast enhanced imaging (DCE).10 So near-simultaneous acquisition of molecular and high-resolution anatomical images with excellent soft-tissue contrast, as well as the utilization of functional MR imaging parameters for the development of diagnostic and intermediate endpoint biomarkers are possible.5

Initially, the MRI component of PET/MRI was primarily utilized for anatomical imaging, similar to PET/CT. While this approach has its advantages, such as superior soft tissue contrast in MRI, relying solely on MRI for anatomical orientation clearly limits the potential of PET/MRI. Consequently, recent studies have focused on the multiparametric capabilities of PET/MRI by incorporating diffusion-weighted sequences and other functional imaging techniques. This integration enables a more comprehensive approach to PET/MR imaging, harnessing its full potential for local primary staging through multiparametric imaging and facilitating a one-stop shop approach for whole-body staging.13 In addition, PET/MRI requires a significantly lower exposure to radiation, with a reduction of 79.7% (range, 72.6%-86.2%), compared to PET/CT.14,15

The integration of PSMA PET/MRI holds the potential for additional value in PCa management. Recent investigations have suggested that PSMA PET/MRI exhibits superior detection efficacy and can significantly impact decision-making processes,16,17 which will be discussed in detail further below.

The objective of medical imaging in PCa patients varies as the disease progresses. During the initial evaluation, the utilization of PSMA PET/MRI allows for the localization of the primary tumor through multiparametric acquisitions, aiding in diagnostic accuracy. As PCa advances, PSMA PET/MRI serves as a valuable tool for comprehensive tumor characterization by providing multiparametric images, enabling a more detailed assessment of the tumor's characteristics. Notably, in cases where metastases undergo dedifferentiation, PSMA PET/MRI plays a crucial role in accurately identifying the presence of metastatic lesions that exhibit faint or no PSMA uptake.

Accurate staging of PCa is crucial for guiding appropriate treatment decisions. In current guidelines PCa is commonly staged using the TNM classification system. Various imaging techniques have been explored for local staging, including TRUS and CT, but their accuracy is limited, especially for advanced cases. The standard TRUS has limitations in accurately detecting PCa,18 and the diagnostic yield of performing additional biopsies on hypoechoic lesions identified through TRUS is found to be negligible.2,19 In contrast, MRI, particularly T2-weighted imaging, has emerged as the most valuable method for local staging. Meta-analysis data shows moderate sensitivity (0.5, 95% CI: 0.49-0.64 and 0.58, 95% CI: 0.47-0.68 and 0.61, 95% CI: 0.54-0.67) but high specificity (0.91, 95% CI: 0.88-0.93 and 0.96, 95% CI: 0.95-0.97 and 0.88, 95% CI: 0.85-0.91) for assessing extraprostatic extension, seminal vesicle invasion, and overall stage T3 using MRI.20,21

As mentioned earlier, when PCa is suspected due to elevated PSA values or suspicious DRE findings and primary imaging modalities like ultrasound and MRI were used, a biopsy for confirmation is conducted.

However, approximately one in five men (20%) undergo unnecessary biopsies due to false positive PSA test results.22 Traditionally, prostate biopsy involves a systematic or image-guided approach. However, recent trials have revolutionized the field by demonstrating that MRI triage provides superior diagnostic accuracy compared to transrectal template biopsy, allowing some men to avoid unnecessary biopsies, and reducing false negative outcomes. The PRECISION trial revealed that MRI, in comparison to 12-core TRUS-guided biopsy, increased the detection rate of clinically significant disease from 26% to 38%, while reducing the identification of clinically insignificant disease from 22% to 9%.23,24

Despite these advancements, the positive predictive value (PPV) of MRI remains relatively low, ranging from 34% to 68%, leading to unnecessary biopsies. Furthermore, MRI fails to detect clinically significant PCa (csPCa) in up to 13% of cases.25

The PRIMARY study has provided prospective confirmation of the high sensitivity of PSMA for detecting ISUP 2 malignancy, demonstrating a 90% sensitivity for clinically significant PCa (csPCa). PSMA and MRI exhibited similar sensitivity and specificity for csPCa, although both modalities showed a relatively high false negative rate, with 17% of csPCa cases being negative on MRI and 10% negative on PSMA. The combination of PSMA + MRI confirmed a synergistic benefit, with both modalities exhibiting higher sensitivity and a higher negative predictive value compared to MRI alone.

The intensity of PSMA uptake was strongly associated with grade group on histopathology, consistent with the known function of PSMA receptors in PCa. This association between grade group and PSMA intensity may have clinical utility: Men with a PSMA SUVmax of 12 or higher were found to have csPCa on biopsy, regardless of MRI findings. Additionally, in men with PI-RADS (Prostate Imaging Reporting and Data System) 4 or 5 lesions, an SUVmax of 9 or higher indicated csPCa with 100% specificity. This suggests that men with intense PSMA uptake and positive MRI findings could potentially avoid confirmatory biopsy and proceed directly to definitive therapy.

The compelling advantage of PSMA in the PRIMARY study was observed in men with negative or equivocal MRI results. On biopsy, 28% of men with PI-RADS 2 lesions and 47% with PI-RADS 3 lesions were found to have csPCa, with PSMA detecting 90% of these malignancies. Furthermore, among men with negative or equivocal MRI findings, 38% had negative PSMA results, and 91% of them had no csPCa on biopsy. Given the high sensitivity of PSMA + MRI in detecting csPCa and the predominantly low-grade characteristics of the missed cancers, it may be acceptable to forego biopsy in men with negative PSMA + MRI findings, in the absence of subsequent concerning PSA kinetics.23

Figure 1, Figure 2, Figure 3 exemplify the advantages of combining PET with mpMRI for PCa diagnosis and management, demonstrating the potential benefits of integrating these modalities in diverse clinical scenarios.

In recent years, the integration of PET/MRI has garnered significant attention as a promising imaging approach for the initial staging and detection of primary disease in PCa. Numerous studies have been dedicated to investigating the diagnostic capabilities and clinical utility of PET/MRI in this context.6,26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40 Integration of PET/MRI demonstrated superior diagnostic value in the localization of PCa when compared to either multiparametric (mp) MRI or PET imaging alone.30,36 For example, in a study conducted by Park et al.34, 68Ga-PSMA-11 PET successfully detected PCa in all 33 patients, while mpMRI using the PI-RADS identified 4 or 5 lesions in 26 patients but missed tumors in 3 individuals.6 Furthermore, the impact of PET/MRI on patient management is a meaningful indicator of its influence during the initial staging of PCa patients. Grubmuller et al.38 found that incorporating PET/MRI in the initial evaluation of PCa patients led to significant changes in therapeutic strategies in at least 30% of cases. So, the integration of PET/MRI offers promising prospects for optimizing individualized care and decision-making during the early stages of PCa diagnosis and treatment planning.6

In cases of BCR after primary therapy, accurate localization of tumor recurrence and its extent are highly relevant for tailored salvage therapies (such as salvage radiation therapy or salvage lymph node dissection). Apart from multiparametric MRI, conventional imaging is rarely performed due to the infrequent identification of both the location and extent of tumor recurrence. So, since the introduction of PSMA-Tracers in the early 2010s there has been a growing utilization of PET imaging due to its enhanced effectiveness in PCa diagnostics.7

For cases of BCR of PCa Evangelista et al. included in their meta-analysis 9 studies with 598 cases. The collective findings of these studies revealed that the recurrent disease detection rate achieved with PET/MRI varied between 54.5% and 97%. Notably, several authors also assessed the detection rate based on PSA categories, showing an increasing trend from low (< 0.2 ng/mL) to high (> 10 ng/mL) PSA levels. For instance, Hope et al.41 and Grubmuller et al.38 reported a detection rate of 58% to 64% for PSA levels <0.5 ng/mL using 68Ga-PSMA-11 PET/MRI, while it reached 100% for PSA > 2.0 ng/mL.

Furthermore, a significant proportion of cases (ranging from 53.2%-74.6%) demonstrated changes in patient management based on PET/MRI findings. The study by Kranzbuhler et al.42 indicated that the inclusion of PET/MRI in the diagnostic workup could potentially lead to changes in RT planning for 39.4% of patients. These results highlight the clinical impact and potential implications of PET/MRI in guiding treatment decisions and enhancing patient care in cases of BCR of PCa.6

As we discussed technical issues of PET/MRI in a review article in more detail,43 we will here point only to the most relevant issues.

In general, to produce quantitative PET data, photon scatter correction (SC) and attenuation correction (AC) has to be taken into account. In PET/MRI, the lack of a signal corresponding to the tissue radiodensity of 511-keV photons raised considerable concerns about its clinical viability, an extrapolation approach was utilized and now, a MR-image-based approach was devised, leveraging the Dixon sequence and image segmentation techniques.44 The Dixon sequence is designed to differentiate fat and water signals which are into attenuation maps with four distinct compartments: fat, soft tissue, background air, and lungs where each compartment is assigned a specific coefficient. However, one notable absence from this approach is bone which potentially is an issue with imaging of the pelvis. This omission is due to the challenging nature of segmenting bone from MR sequences. The extremely short T2 relaxation time of cortical bone causes the MR signal to decay rapidly, making its accurate segmentation cumbersome.

Overall, the developments in the last years have further optimized the Dixon-based AC approach, making it even more effective and valuable in the field of PET/MR imaging.45 As a result, the use of soft tissue LACs for bone AC in PET/MR has become a widely adopted and clinically viable solution, contributing to the success and utility of whole-body oncological PET/MR imaging.46, 47, 48 In order to still improve, an approach to include bone in PET AC was proposed, involving the use of an offline-constructed bone model. This method aims to register the bone model to the Dixon attenuation map, specifically focusing on major bones such as the skull, spine with sacrum, left and right hip, and left and right femur. This presents a promising avenue for addressing potential limitations of solely using soft tissue LACs for bone attenuation in whole-body oncological PET/MR imaging.49 After the commercialization and implementation of the offline-constructed bone model method into the clinical routine, it received mixed feedback from the medical community. Several studies (reported on its performance and impact on clinical interpretations.50, 51, 52 Issues related to inaccurate bone registration and occasional absence of registered bones have been identified, requiring a thorough evaluation of their effects on image quantification using multiple radiotracers—although is a tedious task in routine imaging: inaccurate bone registration can result in misalignments between the bone model and the Dixon attenuation map. This misalignment may lead to errors in estimating bone attenuation, impacting the overall accuracy of PET image quantification. Such inaccuracies can compromise the reliable interpretation of PET findings, particularly in areas where bone plays a significant role in PET signal attenuation such as bone lesions.

In conclusion, despite the advancements in PET/MR attenuation mapping techniques, validation studies are limited due to the absence of a reliable ground truth for comparison. Interestingly, these sophisticated techniques did not yield significant improvements when compared to the simpler and more robust methods of attenuation mapping.52 These results, however, are fortunate because it avoids potential complications in the assessment of effects in serial studies.

In addition to photon attenuation, photon scatter also leads to over- and underestimation of activity concentration, thus impairing PET quantification.53 PET/MRI systems have a small PET ring diameter (around 60 cm) within the MRI bore, leading to a higher chance of random and scattered coincidences—but a reduced likelihood that “tissue free areas” between the body and the detectors are found. To compensate, they use narrower energy windows with comparable energy resolution. This maintains a relatively good scatter fraction, and overall, PET detector performance inside MRI scanners remains largely unaffected. In PET/MRI, additional scatter can occur due to certain PET isotopes that emit prompt gamma rays, such as 82Rb or 68Ga. These higher-energy gammas may get scattered on their way to the detectors, causing them to fall within the PET energy window. As a result, prompt-gamma correction is necessary for PET isotopes with a higher prompt-gamma-branching fraction, ensuring accurate PET image quantification and reducing potential artifacts caused by this phenomenon.54 This is of special importance, as PCa targeting ligands have renal excretion papers with accumulation in kidney and bladder—potential close to the target lesions. Inaccurate SC in PET imaging can lead to the appearance of strong photopenic artifacts known as halo artifacts. These halos are caused by extreme differences in activity concentration between low-uptake background regions and high-uptake organs. As a consequence, the signal amplitude around these organs is drastically reduced, impairing the detectability and quantification of lesions. The presence of these halos can significantly compromise the accuracy and reliability of PET images, making it challenging to interpret and analyze PET data effectively.55 Fortunately, for specific PET tracers like 68Ga-PSMA, the absolute scaling of the most commonly used single-scatter simulation algorithm has been found to be less susceptible to halo artifacts when compared to relative scaling. These findings indicate that adjusting the scatter simulation algorithm's parameters can significantly improve the accuracy and reliability of SC, leading to more reliable and artifact-free PET images for such tracers like.56,57

One of the core limitation of today's PET/MRI system is the high cost for both investment and maintenance. This is (at least) partially related to the industry's initial concept that PET/MRI should compete with PET/CT and accordingly provide whole body application. However, PET/MRI could excel instead as a single organ approach with a lower magnetic field strength, a reduced PET field-of-view and a larger bore diameter. Thus, should there ever be a second iteration for PET/MRI systems, such a less “scan range ambitious” but a targeted and cost effective approach could be very attractive.

New tools in data science promise to improve the analysis of medical images—and data analysis in PCa is no exception here. These methods, based on machine learning and deep learning, are jointly known as radiomics. Several PET/MRI applications have been developed for PCa radiomics, mostly as spin-offs of PET-only or MRI-only radiomics projects. It is still an emerging field, which can be appreciated in its handful of use cases, small patient cohorts and lack of prospective and multicenter studies. An interesting initial application58 was the combination of a support vector machine (SVM) with an unsupervised algorithm to automatically classify tumor voxels in a 16-patient cohort of multiparametric 11C-Ch PET/MRI images. It showed that voxel-wise classification was possible in cancer radiomics, but still with mixed results. Since then, radiomics workflows have been evolving into different applications and growing in predictive power. The prediction of cancer risk and outcome is currently one of the most developed uses of machine learning in the field. One such case was the use of a SVM in manual whole-prostate segmentations of 68Ga-PSMA PET/MRI to predict prostatectomy ISUP Gleason grades as a proxy for primary tumor aggressiveness.59 A rather big, 101-patient cohort was selected to prove that combining PSMA-PET and ADC whole-prostate radiomics can outperform both assuming biopsy ISUP grades and using PET-only or MRI-only metrics, showing a synergetic role of PET and MRI through machine learning. Similar results were later corroborated independently.60 A similar approach was implemented in a 105-patient cohort of 18F-DCFPyL PET/MRI studies.61 In this case, PCa lesions were segmented and a radiomics analysis was successfully performed to find the best set of PET and MRI features to predict low/intermediate vs high risk Gleason groups.

Besides GS, cancer staging, and outcome prediction is also a popular topic. In one application, a group of scientists developed a model to predict cancer staging (TNM), risk groups and patient outcome (PFS) based on 11C-Ch PET/MRI images.62 They showed that using different radiomics regions, including tumor-only and whole-prostate segmentation, can provide complementary information for the prediction. Although PET-only radiomics where used, T2w MRI images allowed a correct prostate and tumor segmentation and were used as part of the risk assessment strategy. Machine learning models combined with pathology studies were also used to predict lesion-wise risk groups.63 In this study, 52 pretreatment 68Ga-PSMA PET/MRI radiomics were successful in predicting patient outcome and high-vs-low lesion and overall risk. In this case, a random forest model outperformed PET-only features like SUVmax and mean SUV.

Besides diagnostic applications, machine learning can also be used to predict treatment outcome.64 In a cohort of 21 advanced PCa patients who underwent 68Ga-PSMA PET/MRI before 177Lu-PSMA therapy, this team showed that a logistic regression model can be fitted on lesion-based radiomics to predict both biochemical response and overall survival of the patients.

Another innovative application was the correlation of genomic index lesions with PET and mpMRI image radiomics. In a proof-of-concept study,65 scientist showed that it is possible to select the most aggressive genomic specimen from a set of repeat biopsies in high-risk PCa patients based only on combined PET and MRI radiomics. Other innovative use cases in PCa management have been previously discussed,66,67 but still no results have been shown.

In summary, although the field of machine learning and AI is evolving fast, clinical uses in PET/MRI for PCa are still an incipient topic in the scientific literature. But the results are encouraging and promise to better harness the synergy between PET and MRI biomarkers.