Study design and patient population

This single-centre, prospective, non-randomized study, (PROSTAGE, registered as NCT03537391) included patients with newly diagnosed PCa of unfavourable intermediate- to high-risk (defined as International Society of Urological Pathology grade group [ISUP GG] ≥ 3, PSA ≥ 20 ng/ml, and/or clinical stage cT ≥ T3). The primary aim of the study was to compare 18F-PSMA-1007 PET/CT, WBMRI with diffusion-weighted imaging (DWI) and conventional imaging modalities – including 99mTc-hydroxymethylene diphosphonate bone scintigraphy, single-photon emission computed tomography-CT (SPECT-CT), and contrast-enhanced CT of the thorax, abdomen and pelvis – for the overall staging of PCa.

The results of nodal (N-staging) and distant metastasis (M-staging) using these imaging modalities have already been reported [9, 10]. This report focuses on intraprostatic tumor detection and T-staging in patients who underwent robot-assisted laparoscopic prostatectomy (RALP).

The key exclusion criteria – prior imaging for PCa metastasis staging or contraindication to MRI – were consistent with those of the PROSTAGE study. For this analysis, an additional exclusion criterion was applied: previous PCa treatment before prostatectomy. All participants underwent WBMRI with DWI and 18F-PSMA-1007 PET/CT prior to RALP. The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics committee of the Hospital District of Southwest Finland.

Imaging modalities

The PET/CT scans were performed using the Discovery MI digital PET/CT system from GE Healthcare (Milwaukee, WI, USA). For attenuation correction, a low-dose CT protocol was used for the transmission scan, featuring a noise index of 30, automatic 3D current modulation (10 to 120 mAs) and 120 kVp. A static emission scan was conducted from the vertex to mid-thigh, covering six bed positions with an acquisition time of two minutes per bed. The sinogram data were corrected for deadtime, decay, and photon attenuation, and reconstructed into a 256 × 256 matrix. Image reconstruction utilized the Q. Clear method, a Bayesian penalized likelihood algorithm for PET, with a β value of 500, incorporating random and scatter corrections. The final in-plane resolution, measured as full-width half-maximum (FWHM), was less than 5 mm.

The WBMRI scans were performed using the Siemens Avanto fit 1.5 T MR system (Siemens Healthcare GmbH, Erlangen, Germany). The imaging protocol consisted of axial T2-weighted fat suppressed (FS) half-Fourier single-shot turbo spin-echo images (HASTE), axial short-tau inversion recovery (STIR) DWI with b-values of 0, 50 and 900 s/mm2, and coronal 3D T1-weighted volumetric interpolated breath-hold examination (VIBE) Dixon sequences. Additionally, STIR DWI images from the level of the pelvis with b-values 0 and 1500 s/mm2 were included.

Imaging interpretation and analysis

PSMA PET/CT and WBMRI images were independently reviewed by two nuclear medicine physicians (8 and 6 years of experience in PSMA PET/CT) and radiologists (12 and 9 years of experience in prostate MRI), respectively. Interpretations were based on clinical expertise and current guidelines [11, 12]. All readers were aware of the diagnosis of PCa but blinded to other clinical, imaging and pathological data.

A PCa-positive lesion on PSMA PET/CT was defined as a focal uptake with activity at least twice that of surrounding prostate tissue, while diffuse and homogeneous increased uptake was not considered PCa-positive. Maximum standardized uptake value (SUVmax) was measured for each lesion. An additional volumetric analysis was performed on index lesions, with a manually defined constraint limited to the prostate gland. Thresholds of 20%, 30%, 40%, and 50% of SUVmax, as well as prostate SUVbackground ×1 and prostate SUVbackground ×2, were assessed. The maximum transaxial diameter of the generated tumour volume was measured for each threshold (Supplementary Figure S1) and compared to the maximum transaxial diameter measured in histology. In cases where PCa affected the entire prostate gland (n = 2), the average prostate background measured from other patients was used as a reference.

In WBMRI, lesions with a score of 3 or higher on the Prostate Imaging Reporting and Data System (PI-RADS) were considered positive for PCa.

In addition to lesion localization, the presence of EPE and SVI were evaluated at the lesion level. In WBMRI, the likelihood of EPE was assessed using a five-point scale, with scores of 4 and 5 regarded as positive for EPE.

Histopathological analysis

The prostatectomy specimens were processed and evaluated following current guidelines [2, 13]. After formalin fixation, the specimens were dissected at 5 mm intervals and embedded in paraffin as whole-mounts. Histological sections, 4 μm thick, were prepared for routine haematoxylin and eosin (H&E) staining and IHC. Lab Vision autostainer (Thermo Fisher Scientific) was used for PSMA staining with a mouse monoclonal PSMA antibody (Dako, clone M3620, 1:100). The slides were digitized with Pannoramic 1000 slide scanner (3DHistec, Hungary) at 20x resolution and analysed with the open-source software QuPath (v. 0.5.0) [14]. Histopathological evaluation was conducted under the supervision of a dedicated uropathologist who was blinded to imaging data. Only PCa lesions with a diameter ≥ 3 mm were included in the analyses. Anatomical location, maximum transaxial diameter, ISUP GG and the presence of EPE, SVI, extensive perineural invasion (defined as > 10 instances with some extending beyond the tumour bulk), ductal carcinoma and cribriform growth pattern were assessed for each lesion. The previously reported 4.3% linear shrinkage of prostatectomy specimens during tissue processing was considered marginal and not taken into account in measuring lesion diameters [15]. The lesion with the highest GG and, in case of multiple lesions with identical GG, the lesion larger in diameter was selected as the index lesion. For IHC slide analysis, lesions were manually annotated, and a cell detection tool was applied to the annotated areas. A supervised object classifier, trained on manually labelled data, was used to identify PCa cells. The software calculated the maximum optical density (ODmax) of cytoplasmic 3,3’-diaminobenzidine (DAB) staining for each identified PCa cell. The mean ODmax was reported for each lesion and for a manually selected hotspot of 20,000 (± 1,000) cells with the strongest staining. “Mean ODmax of cytoplasmic DAB” refers to the per-lesion mean value if not stated otherwise.

Segment-level analysis in imaging and histopathology

To evaluate the anatomical location of each lesion in both imaging and histopathology, the prostate was divided into 12 segments. The basis, mid gland and apex were each subdivided into quadrants: right/left and anterior/posterior. Each of the 12 segments was classified as either positive or negative for the presence of a lesion. Lesions identified by imaging and histopathology were considered to match if they overlapped in at least one segment.

Consensus reading

Following segment-level analysis, a consensus reading was conducted between the pathologist and PET reader 2 to resolve discrepancies in the interpretation of anatomical locations. The pairing of PCa lesions identified in histopathology and PET was manually reviewed case by case and, if multiple PET lesions corresponded to a single histopathological lesion or vice versa, the lesions were grouped accordingly. False-positive PET lesions were re-evaluated in histopathology, and the corresponding tissue blocks were stained for PSMA.

Statistical analysis

Descriptive statistics for categorical variables were reported as frequency and percentage, while continuous variables were summarized using median and interquartile range (IQR). Detection rates were compared between imaging modalities using Pearson’s Chi-squared test with Yate’s continuity correction. Sensitivity, specificity and accuracy were reported with 95% confidence intervals (CI) and compared between modalities using Fisher’s exact test. The area under the curve (AUC) was calculated with 95% CI and compared between modalities using DeLong’s test. Student’s t-test was employed to compare the characteristics of index and non-index lesions, with Levene’s test used to assess equality of variances. Interobserver agreement was evaluated using Cohen’s Kappa. Correlations between SUVmax, ODmax, and ISUP GG were analyzed using Spearman’s rank correlation coefficient. Normality was assessed using the Shapiro-Wilk test. A one-sample Wilcoxon test was used to analyse the difference between lesion diameters in histology and PET/CT. A p-value of less than 0.05 was considered statistically significant. Statistical analyses were performed using IBM SPSS Statistics (version 29.0.0.0) for Windows and R (version 3.6.3).