Prediction of homologous recombination deficiency from routine histology with attention-based multiple instance learning in nine different tumor types.

BACKGROUND: Homologous recombination deficiency (HRD) is recognized as a pan-cancer predictive biomarker that potentially indicates who could benefit from treatment with PARP inhibitors (PARPi). Despite its clinical significance, HRD testing is highly complex. Here, we investigated in a proof-of-concept study whether Deep Learning (DL) can predict HRD status solely based on routine hematoxylin & eosin (H&E) histology images across nine different cancer types. METHODS: We developed a deep learning pipeline with attention-weighted multiple instance learning (attMIL) to predict HRD status from histology images. As part of our approach, we calculated a genomic scar HRD score by combining loss of heterozygosity (LOH), telomeric allelic imbalance (TAI), and large-scale state transitions (LST) from whole genome sequencing (WGS) data of n = 5209 patients across two independent cohorts. The model's effectiveness was evaluated using the area under the receiver operating characteristic curve (AUROC), focusing on its accuracy in predicting genomic HRD against a clinically recognized cutoff value. RESULTS: Our study demonstrated the predictability of genomic HRD status in endometrial, pancreatic, and lung cancers reaching cross-validated AUROCs of 0.79, 0.58, and 0.66, respectively. These predictions generalized well to an external cohort, with AUROCs of 0.93, 0.81, and 0.73. Moreover, a breast cancer-trained image-based HRD classifier yielded an AUROC of 0.78 in the internal validation cohort and was able to predict HRD in endometrial, prostate, and pancreatic cancer with AUROCs of 0.87, 0.84, and 0.67, indicating that a shared HRD-like phenotype occurs across these tumor entities. CONCLUSIONS: This study establishes that HRD can be directly predicted from H&E slides using attMIL, demonstrating its applicability across nine different tumor types.

Dual role of GRHL3 in bladder carcinogenesis depending on histological subtypes.

The effect of grainyhead-like transcription factor 3 (GRHL3) on cancer development depends on the cancer subtypes as shown in tumor entities such as colorectal or oral squamous cell carcinomas. Here, we analyzed the subtype-specific role of GRHL3 in bladder carcinogenesis, comparing common urothelial carcinoma (UC) with squamous bladder cancer (sq-BLCA). We examined GRHL3 mRNA and protein expression in cohorts of patient samples, its prognostic role and its functional impact on tumorigeneses in different molecular and histopathological subtypes of bladder cancer. We showed for GRHL3 a reverse expression in squamous and urothelial bladder cancer subtypes. Stably GRHL3-overexpressing EJ28, J82, and SCaBER in vitro models revealed a tumor-suppressive function in squamous and an oncogenic role in the urothelial cancer cells affecting cell and colony growth, and migratory and invasive capacities. Transcriptomic profiling demonstrated highly subtype-specific GRHL3-regulated expression networks coined by the enrichment of genes involved in integrin-mediated pathways. In SCaBER, loss of ras homolog family member A (RHOA) GTPase activity was demonstrated to be associated with co-regulation of eukaryotic translation initiation factor 4E family member 3 (EIF4E3), a potential tumor suppressor gene. Thus, our data provide for the first time a detailed insight into the role of the transcription factor GRHL3 in different histopathological subtypes of bladder cancer.

Alpha-1 Antitrypsin Augmentation and the Liver Phenotype of Adults With Alpha-1 Antitrypsin Deficiency (Genotype Pi∗ZZ).

BACKGROUND & AIMS: α1-Antitrypsin (AAT) is a major protease inhibitor produced by hepatocytes. The most relevant AAT mutation giving rise to AAT deficiency (AATD), the 'Pi∗Z' variant, causes harmful AAT protein accumulation in the liver, shortage of AAT in the systemic circulation, and thereby predisposes to liver and lung injury. Although intravenous AAT augmentation constitutes an established treatment of AATD-associated lung disease, its impact on the liver is unknown. METHODS: Liver-related parameters were assessed in a multinational cohort of 760 adults with severe AATD (Pi∗ZZ genotype) and available liver phenotyping, of whom 344 received augmentation therapy and 416 did not. Liver fibrosis was evaluated noninvasively via the serum test AST-to-platelet ratio index and via transient elastography-based liver stiffness measurement. Histologic parameters were compared in 15 Pi∗ZZ adults with and 35 without augmentation. RESULTS: Compared with nonaugmented subjects, augmented Pi∗ZZ individuals displayed lower serum liver enzyme levels (AST 71% vs 75% upper limit of normal, P < .001; bilirubin 49% vs 58% upper limit of normal, P = .019) and lower surrogate markers of fibrosis (AST-to-platelet ratio index 0.34 vs 0.38, P < .001; liver stiffness measurement 6.5 vs 7.2 kPa, P = .005). Among biopsied participants, augmented individuals had less pronounced liver fibrosis and less inflammatory foci but no differences in AAT accumulation were noted. CONCLUSIONS: The first evaluation of AAT augmentation on the Pi∗ZZ-related liver disease indicates liver safety of a widely used treatment for AATD-associated lung disease. Prospective studies are needed to confirm the beneficial effects and to demonstrate the potential efficacy of exogenous AAT in patients with Pi∗ZZ-associated liver disease.

Scoring Systems for Immunohistochemistry in Urothelial Carcinoma.

Immunohistochemistry is widely used in diagnostic and scientific analysis of urothelial carcinoma. Objective interpretation of staining results is mandatory for accuracy and comparability in diagnostic and therapeutic patient care as well as research.Herein we summarize and explain standardized microscopic evaluation and scoring approaches for immunohistochemical stainings. We focus on commonly used and generally feasible approaches for different cellular compartments and comment on their utility in diagnostics and research practice.

Mechanisms of organ fibrosis: Emerging concepts and implications for novel treatment strategies.

Fibrosis, or tissue scarring, develops as a pathological deviation from the physiological wound healing response and can occur in various organs such as the heart, lung, liver, kidney, skin, and bone marrow. Organ fibrosis significantly contributes to global morbidity and mortality. A broad spectrum of etiologies can cause fibrosis, including acute and chronic ischemia, hypertension, chronic viral infection (e.g., viral hepatitis), environmental exposure (e.g., pneumoconiosis, alcohol, nutrition, smoking) and genetic diseases (e.g., cystic fibrosis, alpha-1-antitrypsin deficiency). Common mechanisms across organs and disease etiologies involve a sustained injury to parenchymal cells that triggers a wound healing response, which becomes deregulated in the disease process. A transformation of resting fibroblasts into myofibroblasts with excessive extracellular matrix production constitutes the hallmark of disease, however, multiple other cell types such as immune cells, predominantly monocytes/macrophages, endothelial cells, and parenchymal cells form a complex network of profibrotic cellular crosstalk. Across organs, leading mediators include growth factors like transforming growth factor-ß and platelet-derived growth factor, cytokines like interleukin-10, interleukin-13, interleukin-17, and danger-associated molecular patterns. More recently, insights into fibrosis regression and resolution of chronic conditions have deepened our understanding of beneficial, protective effects of immune cells, soluble mediators and intracellular signaling. Further in-depth insights into the mechanisms of fibrogenesis can provide the rationale for therapeutic interventions and the development of targeted antifibrotic agents. This review gives insight into shared responses and cellular mechanisms across organs and etiologies, aiming to paint a comprehensive picture of fibrotic diseases in both experimental settings and in human pathology.

Optical coherence tomography combined with convolutional neural networks can differentiate between intrahepatic cholangiocarcinoma and liver parenchyma ex vivo.

PURPOSE: Surgical resection with complete tumor excision (R0) provides the best chance of long-term survival for patients with intrahepatic cholangiocarcinoma (iCCA). A non-invasive imaging technology, which could provide quick intraoperative assessment of resection margins, as an adjunct to histological examination, is optical coherence tomography (OCT). In this study, we investigated the ability of OCT combined with convolutional neural networks (CNN), to differentiate iCCA from normal liver parenchyma ex vivo. METHODS: Consecutive adult patients undergoing elective liver resections for iCCA between June 2020 and April 2021 (n = 11) were included in this study. Areas of interest from resection specimens were scanned ex vivo, before formalin fixation, using a table-top OCT device at 1310 nm wavelength. Scanned areas were marked and histologically examined, providing a diagnosis for each scan. An Xception CNN was trained, validated, and tested in matching OCT scans to their corresponding histological diagnoses, through a 5 × 5 stratified cross-validation process. RESULTS: Twenty-four three-dimensional scans (corresponding to approx. 85,603 individual) from ten patients were included in the analysis. In 5 × 5 cross-validation, the model achieved a mean F1-score, sensitivity, and specificity of 0.94, 0.94, and 0.93, respectively. CONCLUSION: Optical coherence tomography combined with CNN can differentiate iCCA from liver parenchyma ex vivo. Further studies are necessary to expand on these results and lead to innovative in vivo OCT applications, such as intraoperative or endoscopic scanning.

Direct prediction of Homologous Recombination Deficiency from routine histology in ten different tumor types with attention-based Multiple Instance Learning: a development and validation study.

Background: Homologous Recombination Deficiency (HRD) is a pan-cancer predictive biomarker that identifies patients who benefit from therapy with PARP inhibitors (PARPi). However, testing for HRD is highly complex. Here, we investigated whether Deep Learning can predict HRD status solely based on routine Hematoxylin & Eosin (H&E) histology images in ten cancer types. Methods: We developed a fully automated deep learning pipeline with attention-weighted multiple instance learning (attMIL) to predict HRD status from histology images. A combined genomic scar HRD score, which integrated loss of heterozygosity (LOH), telomeric allelic imbalance (TAI) and large-scale state transitions (LST) was calculated from whole genome sequencing data for n=4,565 patients from two independent cohorts. The primary statistical endpoint was the Area Under the Receiver Operating Characteristic curve (AUROC) for the prediction of genomic scar HRD with a clinically used cutoff value. Results: We found that HRD status is predictable in tumors of the endometrium, pancreas and lung, reaching cross-validated AUROCs of 0.79, 0.58 and 0.66. Predictions generalized well to an external cohort with AUROCs of 0.93, 0.81 and 0.73 respectively. Additionally, an HRD classifier trained on breast cancer yielded an AUROC of 0.78 in internal validation and was able to predict HRD in endometrial, prostate and pancreatic cancer with AUROCs of 0.87, 0.84 and 0.67 indicating a shared HRD-like phenotype is across tumor entities. Conclusion: In this study, we show that HRD is directly predictable from H&E slides using attMIL within and across ten different tumor types.

Characterisation of tumour-immune phenotypes and PD-L1 positivity in squamous bladder cancer.

AIMS: Immune checkpoint inhibitor (ICI) therapy has become a viable treatment strategy in bladder cancer. However, treatment responses vary, and improved biomarkers are needed. Crucially, the characteristics of immune cells remain understudied especially in squamous differentiated bladder cancer (sq-BLCA). Here, we quantitatively analysed the tumour-immune phenotypes of sq-BLCA and correlated them with PD-L1 expression and FGFR3 mutation status. METHODS: Tissue microarrays (TMA) of n = 68 non-schistosomiasis associated pure squamous cell carcinoma (SCC) and n = 46 mixed urothelial carcinoma with squamous differentiation (MIX) were subjected to immunohistochemistry for CD3, CD4, CD8, CD56, CD68, CD79A, CD163, Ki67, perforin and chloroacetate esterase staining. Quantitative image evaluation was performed via digital image analysis. RESULTS: Immune infiltration was generally higher in stroma than in tumour regions. B-cells (CD79A) were almost exclusively found in stromal areas (sTILs), T-lymphocytes and macrophages were also present in tumour cell areas (iTILs), while natural killer cells (CD56) were nearly missing in any area. Tumour-immune phenotype distribution differed depending on the immune cell subset, however, hot tumour-immune phenotypes (high density of immune cells in tumour areas) were frequently found for CD8 + T-cells (33%), especially perforin + lymphocytes (52.2%), and CD68 + macrophages (37.6%). Perforin + CD8 lymphocytes predicted improved overall survival in sq-BLCA while high PD-L1 expression (CPS ≥ 10) was significantly associated with higher CD3 + , CD8 + and CD163 + immune cell density and high Ki67 (density) of tumour cells. Furthermore, PD-L1 expression was positively associated with CD3 + /CD4 + , CD3 + /CD8 + and CD68 + /CD163 + hot tumour-immune phenotypes. FGFR3 mutation status was inversely associated with CD8 + , perforin + and CD79A + lymphocyte density. CONCLUSIONS: Computer-based image analysis is an efficient tool to analyse immune topographies in squamous bladder cancer. Hot tumour-immune phenotypes with strong PD-L1 expression might pose a promising subgroup for clinically successful ICI therapy in squamous bladder cancer and warrant further investigation.

Clinical sequencing identifies potential actionable alterations in a high rate of urachal and primary bladder adenocarcinomas.

OBJECTIVE: Administration of targeted therapies provides a promising treatment strategy for urachal adenocarcinoma (UrC) or primary bladder adenocarcinoma (PBAC); however, the selection of appropriate drugs remains difficult. Here, we aimed to establish a routine compatible methodological pipeline for the identification of the most important therapeutic targets and potentially effective drugs for UrC and PBAC. METHODS: Next-generation sequencing, using a 161 cancer driver gene panel, was performed on 41 UrC and 13 PBAC samples. Clinically relevant alterations were filtered, and therapeutic interpretation was performed by in silico evaluation of drug-gene interactions. RESULTS: After data processing, 45/54 samples passed the quality control. Sequencing analysis revealed 191 pathogenic mutations in 68 genes. The most frequent gain-of-function mutations in UrC were found in KRAS (33%), and MYC (15%), while in PBAC KRAS (25%), MYC (25%), FLT3 (17%) and TERT (17%) were recurrently affected. The most frequently affected pathways were the cell cycle regulation, and the DNA damage control pathway. Actionable mutations with at least one available approved drug were identified in 31/33 (94%) UrC and 8/12 (67%) PBAC patients. CONCLUSIONS: In this study, we developed a data-processing pipeline for the detection and therapeutic interpretation of genetic alterations in two rare cancers. Our analyses revealed actionable mutations in a high rate of cases, suggesting that this approach is a potentially feasible strategy for both UrC and PBAC treatments.

Effective Radiosensitization of Bladder Cancer Cells by Pharmacological Inhibition of DNA-PK and ATR.

This study aims at analyzing the impact of the pharmacological inhibition of DNA damage response (DDR) targets (DNA-PK and ATR) on radiosensitization of bladder cancer cell lines of different molecular/histological subtypes. Applying DNA-PK (AZD7648) and ATR (Ceralasertib) inhibitors on SCaBER, J82 and VMCUB-1 bladder cancer cell lines, we revealed sensitization upon ionizing radiation (IR), i.e., the IC50 for each drug shifted to a lower drug concentration with increased IR doses. In line with this, drug exposure retarded DNA repair after IR-induced DNA damage visualized by a neutral comet assay. Western blot analyses confirmed specific inhibition of targeted DDR pathways in the analyzed bladder cancer cell lines, i.e., drugs blocked DNA-PK phosphorylation at Ser2056 and the ATR downstream mediator CHK1 at Ser317. Interestingly, clonogenic survival assays indicated a cell-line-dependent synergism of combined DDR inhibition upon IR. Calculating combined index (CI) values, with and without IR, according to the Chou-Talalay method, confirmed drug- and IR-dose-specific synergistic CI values. Thus, we provide functional evidence that DNA-PK and ATR inhibitors specifically target corresponding DDR pathways retarding the DNA repair process at nano-molar concentrations. This, in turn, leads to a strong radiosensitizing effect and impairs the survival of bladder cancer cells.

LINC00152 Drives a Competing Endogenous RNA Network in Human Hepatocellular Carcinoma.

Genomic and epigenomic studies revealed dysregulation of long non-coding RNAs in many cancer entities, including liver cancer. We identified an epigenetic mechanism leading to upregulation of the long intergenic non-coding RNA 152 (LINC00152) expression in human hepatocellular carcinoma (HCC). Here, we aimed to characterize a potential competing endogenous RNA (ceRNA) network, in which LINC00152 exerts oncogenic functions by sponging miRNAs, thereby affecting their target gene expression. Database and gene expression data of human HCC were integrated to develop a potential LINC00152-driven ceRNA in silico. RNA immunoprecipitation and luciferase assay were used to identify miRNA binding to LINC00152 in human HCC cells. Functionally active players in the ceRNA network were analyzed using gene editing, siRNA or miRNA mimic transfection, and expression vectors in vitro. RNA expression in human HCC in vivo was validated by RNA in situ hybridization. Let-7c-5p, miR-23a-3p, miR-125a-5p, miR-125b-5p, miR-143a-3p, miR-193-3p, and miR-195-5p were detected as new components of the potential LINC00152 ceRNA network in human HCC. LINC00152 was confirmed to sponge miR143a-3p in human HCC cell lines, thereby limiting its binding to their respective target genes, like KLC2. KLC2 was identified as a central mediator promoting pro-tumorigenic effects of LINC00152 overexpression in HCC cells. Furthermore, co-expression of LINC00152 and KLC2 was observed in human HCC cohorts and high KLC2 expression was associated with shorter patient survival. Functional assays demonstrated that KLC2 promoted cell proliferation, clonogenicity and migration in vitro. The LINC00152-miR-143a-3p-KLC2 axis may represent a therapeutic target in human HCC.

PD-1+ T-Cells Correlate with Nerve Fiber Density as a Prognostic Biomarker in Patients with Resected Perihilar Cholangiocarcinoma.

BACKGROUND AND AIMS: Perihilar cholangiocarcinoma (pCCA) is a hepatobiliary malignancy, with a dismal prognosis. Nerve fiber density (NFD)-a novel prognostic biomarker-describes the density of small nerve fibers without cancer invasion and is categorized into high numbers and low numbers of small nerve fibers (high vs low NFD). NFD is different than perineural invasion (PNI), defined as nerve fiber trunks invaded by cancer cells. Here, we aim to explore differences in immune cell populations and survival between high and low NFD patients. APPROACH AND RESULTS: We applied multiplex immunofluorescence (mIF) on 47 pCCA patients and investigated immune cell composition in the tumor microenvironment (TME) of high and low NFD. Group comparison and oncological outcome analysis was performed. CD8+PD-1 expression was higher in the high NFD than in the low NFD group (12.24 × 10-6 vs. 1.38 × 10-6 positive cells by overall cell count, p = 0.017). High CD8+PD-1 expression was further identified as an independent predictor of overall (OS; Hazard ratio (HR) = 0.41; p = 0.031) and recurrence-free survival (RFS; HR = 0.40; p = 0.039). Correspondingly, the median OS was 83 months (95% confidence interval (CI): 18-48) in patients with high CD8+PD-1+ expression compared to 19 months (95% CI: 5-93) in patients with low CD8+PD-1+ expression (p = 0.018 log rank). Furthermore, RFS was significantly lower in patients with low CD8+PD-1+ expression (14 months (95% CI: 6-22)) compared to patients with high CD8+PD-1+ expression (83 months (95% CI: 17-149), p = 0.018 log rank). CONCLUSIONS: PD-1+ T-cells correlate with high NFD as a prognostic biomarker and predict good survival; the biological pathway needs to be investigated.

Benchmarking weakly-supervised deep learning pipelines for whole slide classification in computational pathology.

Artificial intelligence (AI) can extract visual information from histopathological slides and yield biological insight and clinical biomarkers. Whole slide images are cut into thousands of tiles and classification problems are often weakly-supervised: the ground truth is only known for the slide, not for every single tile. In classical weakly-supervised analysis pipelines, all tiles inherit the slide label while in multiple-instance learning (MIL), only bags of tiles inherit the label. However, it is still unclear how these widely used but markedly different approaches perform relative to each other. We implemented and systematically compared six methods in six clinically relevant end-to-end prediction tasks using data from N=2980 patients for training with rigorous external validation. We tested three classical weakly-supervised approaches with convolutional neural networks and vision transformers (ViT) and three MIL-based approaches with and without an additional attention module. Our results empirically demonstrate that histological tumor subtyping of renal cell carcinoma is an easy task in which all approaches achieve an area under the receiver operating curve (AUROC) of above 0.9. In contrast, we report significant performance differences for clinically relevant tasks of mutation prediction in colorectal, gastric, and bladder cancer. In these mutation prediction tasks, classical weakly-supervised workflows outperformed MIL-based weakly-supervised methods for mutation prediction, which is surprising given their simplicity. This shows that new end-to-end image analysis pipelines in computational pathology should be compared to classical weakly-supervised methods. Also, these findings motivate the development of new methods which combine the elegant assumptions of MIL with the empirically observed higher performance of classical weakly-supervised approaches. We make all source codes publicly available at https://github.com/KatherLab/HIA, allowing easy application of all methods to any similar task.

Heterogenous NECTIN4 expression in urothelial high-risk non-muscle-invasive bladder cancer.

High-grade non-muscle-invasive bladder cancer (HG NMIBC) patients are at high risk (HR) of progression to muscle-invasion. Bladder-preserving therapies for this patient subgroup are limited, and additional treatments are desirable. Recently, enfortumab vedotin, targeting cancer-associated NECTIN4, has been approved for the treatment of advanced urothelial carcinoma. However, data on the expression of NECTIN4 and its therapeutic potential for HR NMIBC are scarce. Here, NECTIN4 was immunohistochemically analyzed in urothelial HG NMIBC by studying cohorts of carcinoma in situ (CIS)/T1HG (N = 182 samples), HG papillary tumors from mixed-grade lesions (mixed TaHG) (N = 87) and papillary HG tumors without a history of low-grade disease (pure TaHG/T1HG) (N = 98) from overall 225 patients. Moreover, inter-lesional NECTIN4 heterogeneity in multifocal HG NMIBC tumors was determined. A high prevalence of NECTIN4 positivity was noted across HG NMIBC subgroups (91%, N = 367 samples), with 77% of samples showing moderate/strong expression. Heterogenous NECTIN4 levels were observed between HG NMIBC subgroups: non-invasive areas of CIS/T1HG and pure TaHG/T1HG samples showed NECTIN4 positivity in 96% and 99%, with 88% and 83% moderate/strong expressing specimens, respectively, whereas significantly lower NECTIN4 levels were detected in mixed TaHG lesions (72% positivity, 48% of samples with moderate/strong NECTIN4 expression). Moreover, higher NECTIN4 heterogeneity was observed in patients with multifocal mixed TaHG tumors (22% of patients) compared to patients with multifocal CIS/T1HG and pure TaHG/T1HG tumors (9% and 5%). Taken together, NECTIN4-directed antibody-drug conjugates might be promising for the treatment of HR NMIBC patients, especially for those exhibiting CIS/T1HG and pure TaHG/T1HG tumors without a history of low-grade disease.

Artificial Intelligence-based Detection of FGFR3 Mutational Status Directly from Routine Histology in Bladder Cancer: A Possible Preselection for Molecular Testing?

BACKGROUND: Fibroblast growth factor receptor (FGFR) inhibitor treatment has become the first clinically approved targeted therapy in bladder cancer. However, it requires previous molecular testing of each patient, which is costly and not ubiquitously available. OBJECTIVE: To determine whether an artificial intelligence system is able to predict mutations of the FGFR3 gene directly from routine histology slides of bladder cancer. DESIGN, SETTING, AND PARTICIPANTS: We trained a deep learning network to detect FGFR3 mutations on digitized slides of muscle-invasive bladder cancers stained with hematoxylin and eosin from the Cancer Genome Atlas (TCGA) cohort (n = 327) and validated the algorithm on the "Aachen" cohort (n = 182; n = 121 pT2-4, n = 34 stroma-invasive pT1, and n = 27 noninvasive pTa tumors). OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: The primary endpoint was the area under the receiver operating curve (AUROC) for mutation detection. Performance of the deep learning system was compared with visual scoring by an uropathologist. RESULTS AND LIMITATIONS: In the TCGA cohort, FGFR3 mutations were detected with an AUROC of 0.701 (p < 0.0001). In the Aachen cohort, FGFR3 mutants were found with an AUROC of 0.725 (p < 0.0001). When trained on TCGA, the network generalized to the Aachen cohort, and detected FGFR3 mutants with an AUROC of 0.625 (p = 0.0112). A subgroup analysis and histological evaluation found highest accuracy in papillary growth, luminal gene expression subtypes, females, and American Joint Committee on Cancer (AJCC) stage II tumors. In a head-to-head comparison, the deep learning system outperformed the uropathologist in detecting FGFR3 mutants. CONCLUSIONS: Our computer-based artificial intelligence system was able to detect genetic alterations of the FGFR3 gene of bladder cancer patients directly from histological slides. In the future, this system could be used to preselect patients for further molecular testing. However, analyses of larger, multicenter, muscle-invasive bladder cancer cohorts are now needed in order to validate and extend our findings. PATIENT SUMMARY: In this report, a computer-based artificial intelligence (AI) system was applied to histological slides to predict genetic alterations of the FGFR3 gene in bladder cancer. We found that the AI system was able to find the alteration with high accuracy. In the future, this system could be used to preselect patients for further molecular testing.