Investigation of histopathology slides by pathologists is an indispensable component of the routine diagnosis of cancer. Artificial intelligence (AI) has the potential to enhance diagnostic accuracy, improve efficiency, and patient outcomes in clinical pathology. However, variations in tissue preparation, staining protocols, and histopathology slide digitization could result in over-fitting of deep learning models when trained on the data from only one center, thereby underscoring the necessity to generalize deep learning networks for multi-center use. Several techniques, including the use of grayscale images, color normalization techniques, and Adversarial Domain Adaptation (ADA) have been suggested to generalize deep learning algorithms, but there are limitations to their effectiveness and discriminability. Convolutional Neural Networks (CNNs) exhibit higher sensitivity to variations in the amplitude spectrum, whereas humans predominantly rely on phase-related components for object recognition. As such, we propose Adversarial fourIer-based Domain Adaptation (AIDA) which applies the advantages of a Fourier transform in adversarial domain adaptation. We conducted a comprehensive examination of subtype classification tasks in four cancers, incorporating cases from multiple medical centers. Specifically, the datasets included multi-center data for 1113 ovarian cancer cases, 247 pleural cancer cases, 422 bladder cancer cases, and 482 breast cancer cases. Our proposed approach significantly improved performance, achieving superior classification results in the target domain, surpassing the baseline, color augmentation and normalization techniques, and ADA. Furthermore, extensive pathologist reviews suggested that our proposed approach, AIDA, successfully identifies known histotype-specific features. This superior performance highlights AIDA’s potential in addressing generalization challenges in deep learning models for multi-center histopathology datasets.

Platinum-based neoadjuvant chemotherapy prior to radical cystectomy is the preferred treatment for muscle-invasive bladder cancer despite modest survival benefit and significant associated toxicities. Here, we profile the global proteome of muscle-invasive bladder cancers pre- and post-neoadjuvant chemotherapy treatment using archival formalin-fixed paraffin-embedded tissue. We identify four pre-neoadjuvant chemotherapy proteomic clusters with distinct biology and response to therapy and integrate these with transcriptomic subtypes and immunohistochemistry. We observe proteomic plasticity post-neoadjuvant chemotherapy that is associated with increased extracellular matrix and reduced keratinisation compared to pre-neoadjuvant chemotherapy. Post-neoadjuvant chemotherapy clusters appear to be differentially enriched for druggable proteins. For example, MTOR and PARP are over-expressed at the protein level in tumours identified as neuronal-like. In addition, we determine that high intra-tumoural proteome heterogeneity in pre-neoadjuvant chemotherapy tissue is associated with worse prognosis. Our work highlights aspects of muscle-invasive bladder cancer biology associated with clinical outcomes and suggests biomarkers and therapeutic targets based on proteomic clusters.

RNA control spike-in mix was not added in this instance. This correction has been made to the HTML and PDF versions of the article.

Background: Immunohistochemistry (IHC) plays a central role in subtyping of muscle-invasive bladder cancer (MIBC), yet conventional semi-quantitative scoring lacks objectivity and scalability. Automated digital pathology offers potential solutions, but requires robust, marker-specific validation against expert consensus scoring. Methods: We developed and internally validated an automated digital pathology pipeline for continuous IHC H-score quantification using QuPath (v0.6.0-arm64). Tissue microarrays (TMAs) were generated from transurethral resection of bladder tumor (TURBT) specimens from a cohort of patients with MIBC treated at Vancouver General Hospital. Cell detection was performed using StarDist (v0.9), followed by automated intensity-based H-score calculation for four epithelial IHC marker stains (CK14, CK20, CK5/6, and Uroplakin II). H-scoring was then restricted to tumor epithelium by object-level classification using a supervised tumor/non-tumor classifier trained on pathologist-reviewed annotations. Automated scores were compared with consensus scores from three blinded pathologists using Pearson correlation, linear regression, intraclass correlation coefficients (ICC), and Bland–Altman analysis. Results: Automated H-scores demonstrated strong agreement with pathologist consensus across all four markers. CK14 showed near-perfect agreement (ICC ≈ 0.99) with minimal bias and narrow limits of agreement. CK20 also showed high agreement (ICC ≈ 0.95). CK5/6 and Uroplakin II demonstrated slightly lower agreement (ICC ≈ 0.92 to 0.93) with mild proportional bias. Across markers, the automated pipeline preserved a broad H-score range, with range ratios of 0.96 to 0.99. Conclusions: This study establishes a robust, methods-forward pipeline for automated continuous IHC H-scoring in MIBC. The internally validated framework provides a scalable foundation for external cohort testing and future clinical outcome-associated biomarker analyses.