Chris H. Bangma

Background and objective: High⍰risk non⍰muscle⍰invasive bladder cancer (HR⍰NMIBC) is treated with transurethral resection and intravesical BCG instillations, yet ≈50% recur and 20% progress to invasive disease. Although molecular subtyping, e.g., BCG-response-subtype (BRS), is associated with progression risk and may aid risk stratification, yet is costly and time-consuming. Intratumoral heterogeneity complicates accurate subtyping. To address these challenges, we developed a deep-learning model that predicts BRS from routine hematoxylin-eosin-stained images. We verified the model’s area-by-area predictions against tissue-level gene-expression maps. Methods and participants: Hematoxylin-eosin-stained images from 231 HR-NMIBC patients with known BRS were used to develop a deep-learning model through cross-validation, then validated in 83 independent samples. The model’s spatial predictions were assessed using spatial transcriptomics to map gene expression to tissue locations in five HR-NMIBC tumors. Outcome measurements and statistical analysis: Discriminative ability for BRS3 vs. BRS1/2 was measured by AUC. Spatial alignment was assessed by calculating Pearson and Spearman correlation coefficients between model predictions and BRS fractions; significance was assessed through permutation analysis. Key findings and limitations: The trained algorithm achieved AUC of 0.79 (development) and 0.71 (external) to detect BRS3 vs BRS1/2. Tile-level correlation between model output and molecular labels was significant (Pearson r = 0.33–0.44; p ≤ 0.002). Limitations include retrospective sampling and limited spatial transcriptomic cases. Conclusions and clinical implications: Our trained algorithm showed potential to stratify HR⍰NMIBC patients by clinically relevant BCG⍰response subtypes using routine hematoxylin-eosin-stained images and showed predicted spatial heterogeneity comparable to molecular profiling. Prospective validation is required before any clinical implementation. Patient summary: Standard pathology images contain hidden details related to tumor’s molecular subtype. We trained an AI model to read these routine images and identify specific bladder cancer subtypes associated with poor response to BCG therapy. This approach may help reveal molecular subtype-associated information from routine pathology images, without additional laboratory procedures.

There is a strong need for nanobodies that target novel cancer-associated antigens to advance radioligand imaging and antibody-based therapeutics. In this study, we investigated whether non-targeted llama immunization using tumor cells, combined with non-targeted phage-display panning of human cell lines, could yield nanobodies specific to Prostate-Specific Membrane Antigen (PSMA). Nanobody selection using both classical three-round PSMA negative–positive panning and single-round panning of cell lines (positive or negative) for PSMA showed clear enrichment for PSMA binders in both strategies. Using shRNA knockdown, flow cytometry, cell-ELISA, immunohistochemistry and structural modeling and docking, we confirmed the PSMA-targeting of selected nanobodies. Two distinct epitopes were predicted to be bound by nanobodies PSMANb9 and A7 (JVZ-007), and this was corroborated by epitope competition assays. These findings support the feasibility of non-targeted immunization and panning strategies for isolating antigen-targeting cancer nanobodies.

Current guidelines recommend pathologists to report percentage Gleason pattern 4 (GP4%) in Gleason score (GS) 7 prostate cancer (PCa) biopsies. However, it is unspecified whether the highest or overall GP4% should be reported. This study aims to clarify which quantification method correlates best with radical prostatectomy (RP) pathology. This study included 308 men with the highest GS 3 + 4 = 7, 4 + 3 = 7, or 4 + 4 = 8 on centrally revised systematic and/or targeted biopsies who underwent RP between 2018 and 2022. The highest and overall biopsy GP4% were compared with RP GP4% using Spearman’s rank correlation coefficient and adverse pathology (AP) (pT-stage ≥ T3, GS ≥ 4 + 3 = 7 and/or pN1) using multivariable logistic regression models adjusted for clinical tumor stage, prostate specific antigen (PSA), percentage of tumor positive biopsies, biopsy modality (systematic/targeted/both), and cribriform pattern. Both quantification methods correlated with RP GP4% (both rho = 0.59), and no significant difference was found between them ( p = 0.78). On multivariable analyses, both GP4% quantification methods were significantly associated with AP (per 10% increase, highest GP4% odds ratio [OR] 1.26 [95% CI 1.14–1.39], overall GP4% OR 1.38 [95% CI 1.22–1.58], both p < 0.001). The area under the curve (AUC) was slightly better for overall (0.78 [95% CI 0.73–0.83]) than the highest GP4% (0.76 [95% CI 0.71–0.81], p = 0.041). This study found that the highest and overall biopsy GP4% both correlated with RP GP4%. Although the discriminative performance of the highest and overall GP4% was comparable with respect to AP at RP, the overall GP4% statistically slightly outperformed the highest GP4%. Including the overall GP4% may have added value in risk stratification and clinical decision-making in a subset of PCa patients.

Aims: To investigate the surgical margin status in patients with prostate cancer who underwent robot‐assisted radical prostatectomy (RARP) with intraoperative neurovascular structure‐adjacent frozen‐section analysis (NeuroSAFE) and evaluate differences compared to patients who underwent radical prostatectomy without NeuroSAFE.Patients and Methods: Between September 2018 and January 2021, 962 patients underwent centralized RARP with NeuroSAFE. A secondary resection was performed in case of a positive surgical margin (PSM) on intraoperative frozen section (IFS) analysis to convert a PSM into a negative surgical margin (NSM). A retrospective cohort consisted of 835 patients who had undergone radical prostatectomy in a tertiary centre without NeuroSAFE between January 2000 and December 2017. We performed multivariable logistic regression to evaluate differences in risk of PSM between cohorts after controlling for clinicopathological variables.Results: Patients operated with NeuroSAFE in the centralized clinic had 29% PSM at a definitive pathological RP examination. The median cumulative length of definitive PSM was 1.1 mm (interquartile range: 0.4–3.8). Among 275 men with PSM, 136 (49%) had a cumulative length ≤1 mm and 198 (72%) ≤3 mm. After controlling for PSA, Grade group, cribriform pattern, pT‐stage, and pN‐stage, patients treated in the centralized clinic with NeuroSAFE had significantly lower odds on PSM (odds ratio [OR]: 0.70, 95% confidence interval [CI]: 0.56–0.88; P = 0.002), PSM length >1 mm (OR: 0.14, 95% CI: 0.09–0.22; P < 0.001), and >3 mm (OR: 0.21, 95% CI: 0.14–0.30; P < 0.001).Conclusion: This study provides a detailed overview of surgical margin status in a centralized RP NeuroSAFE cohort. Centralization with NeuroSAFE was associated with lower PSM rates and significantly shorter PSM cumulative lengths, indicating improved control of surgical margin status.

The computer-aided analysis of prostate multiparametric MRI (mpMRI) could improve significant-prostate-cancer (PCa) detection. Various deep-learning- and radiomics-based methods for significant-PCa segmentation or classification have been reported in the literature. To be able to assess the generalizability of the performance of these methods, using various external data sets is crucial. While both deep-learning and radiomics approaches have been compared based on the same data set of one center, the comparison of the performances of both approaches on various data sets from different centers and different scanners is lacking. The goal of this study was to compare the performance of a deep-learning model with the performance of a radiomics model for the significant-PCa diagnosis of the cohorts of various patients. We included the data from two consecutive patient cohorts from our own center (n = 371 patients), and two external sets of which one was a publicly available patient cohort (n = 195 patients) and the other contained data from patients from two hospitals (n = 79 patients). Using multiparametric MRI (mpMRI), the radiologist tumor delineations and pathology reports were collected for all patients. During training, one of our patient cohorts (n = 271 patients) was used for both the deep-learning- and radiomics-model development, and the three remaining cohorts (n = 374 patients) were kept as unseen test sets. The performances of the models were assessed in terms of their area under the receiver-operating-characteristic curve (AUC). Whereas the internal cross-validation showed a higher AUC for the deep-learning approach, the radiomics model obtained AUCs of 0.88, 0.91 and 0.65 on the independent test sets compared to AUCs of 0.70, 0.73 and 0.44 for the deep-learning model. Our radiomics model that was based on delineated regions resulted in a more accurate tool for significant-PCa classification in the three unseen test sets when compared to a fully automated deep-learning model.