Prostate-specific membrane antigen (PSMA) is upregulated in high-grade glioma (HGG). This study aimed, firstly, to establish dosimetry for the radionuclide [68Ga]-PSMA-11 in HGG patients; secondly, to determine theoretical tumour doses for [177Lu]-PSMA, a potential therapeutic radionuclide; thirdly, to assess PSMA immunohistochemistry in targeted intra-operative HGG biopsies. Three HGG patients underwent PET-MRI after injection of 185 MBq [68Ga]-PSMA-11. Targeted intra-operative HGG biopsies were immunostained for PSMA and endothelium (CD34). There was durable, heterogeneous uptake of [ 68Ga]-PSMA-11 with moderate SUVmax (4.2-5.0) and high TBR (60-183). [68Ga]-PSMA-11 delivered a tumour dose of 0.01-0.03 mGy/MBq corresponding to 0.38-1.10 mGy/MBq for [177Lu]-PSMA. PSMA staining was predominantly seen in necrotic/proliferating CD34-positive cells. HGGs exhibited moderate and heterogeneous [68Ga]-PSMA-11 uptake, corresponding to a theoretical [177Lu] tumour dose lower than conventional external beam radiotherapy but within the range used for theranostic treatment in prostate cancer. PSMA staining was most prevalent in regions of tumour with necrotic/proliferating endothelium.

Abstract Non-small cell lung cancer (NSCLC) with its rapid growth and early metastasis onset, represents the most common cause of cancer-related deaths worldwide. Current treatments offer limited long-term survival, necessitating novel approaches. Immunotherapy, specifically anti-PD1 and anti-PD-L1, has transformed NSCLC treatment, but only a small percentage of patients respond. There is an unmet need to predict immuno-therapy susceptibilities, achieve a cancer patient risk stratification and identify targeted therapies. Patient-derived organoids (PDOs) are in vitro 3D structures that recapitulate the complexity of the tumours from which they derived, showing a great potential as preclinical model for drug screening and tailored treatment. We used 12 PDO models established from chemotherapy-naïve high-risk NSCLC patients tissues collected at the Guy’s Hospital, London. Patients’ follow-up was performed for more than 24 months post resection/biopsy/surgery. PDOs were treated with cisplatin, then co-cultured with pre-activated immune cells in the presence of IO drugs and subjected to multiple assays including imaging, flow cytometry, genomic and transcriptomic analysis and proteomic from exosome isolation. This multi-omics approach allows for simultaneous analysis of different biological parameters before and after treatment to understand tumour cell interaction with immune cells and drug-induced changes. According to patients’ clinical response to standard of care treatment, PDOs were classified as responder (R) and not responder (NR). Cisplatin IC50, elaborated from PDOs cytotoxicity analysis, correlated with R and NR dichotomy, predicting the response status of the patients in vivo. When PDOs were cocultured with PBMC, counterintuitively, we observed a higher immune cells infiltration after cisplatin treatment in the NR-PDOs. Neither baseline CD45 infiltration nor the composition of infiltrating PBMC differed between NR and R-PDOs. However, in the baseline condition, the CD8pos cells infiltrating NR-PDOs presented a predominantly exhaustion profile. Transcriptomic analysis of untreated PDOs revealed higher activation of inflammatory response signalling in the NR-PDOs compared to responders, and the same signalling upregulation was identify from spatial transcriptomic analysis on primary tumour tissues derived from NR patients. Proteomic analysis of extracellular vesicles cargo also identified different protein content in R vs NR-PDO supporting the involvement of EV in the crosstalk between PDO and immune cells. We found that PDOs maintain the transcriptomic profile of the tissue they derived from with different activation of specific signalling that contributes to the creation of an inflammatory microenvironment and correlates with drug response status. Citation Format: Anna Pasto, Halh Al-Serori, Maria Fankhaenel, Zeinab Mokhtari, Debayan Mukherjee, Ruben Drews, Paul Barber, Jessica Davis, Lena Wedeken, Veronika Yankova, Jürgen Loskutov, James Monypenny, Nikunjkumar Prakashbhai Patel, Mint Htun, Michael Finn, Susan Ndagire, Eleni Karapanagiotou, Edmund Moon, Cheryl Gillett, Andrea Bille, Tony Ng. NSCLC patient-derived organoids recapitulate tissue signalling and impairment of infiltrated immune cell activation predicting patients’ response to therapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1386.

Cancer research emphasises early detection, yet quantitative methods for normal tissue analysis remain limited. Digitised haematoxylin and eosin (H&E)-stained slides enable computational histopathology, but artificial intelligence (AI)-based analysis of normal breast tissue (NBT) in whole slide images (WSIs) remains scarce. We curated 70 WSIs of NBTs from multiple sources and cohorts with pathologist-guided manual annotations of epithelium, stroma, and adipocytes (https://github.com/cancerbioinformatics/OASIS). We developed robust convolutional neural network (CNN)-based, patch-level classification models, named NBT-Classifiers, to tessellate and classify NBTs at different scales. Across three external cohorts, NBT-Classifiers trained on 128 × 128 µm and 256 × 256 µm patches achieved AUCs of 0.98–1.00. The model learned independent normal features different from those of precancerous and cancerous epithelium, which were further visualised using two explainable AI techniques. When integrated into an end-to-end preprocessing pipeline, NBT-Classifiers facilitate efficient downstream analysis within peri-lobular regions. NBT-Classifiers provide robust compartment-specific analytical tools and enhance our understanding of NBT appearances, which serve as valuable reference points for identifying premalignant changes and guiding early breast cancer prevention strategies.

The immunosuppressive transmembrane protein PD-L1 was shown to traffic via the multivesicular body (MVB) and to be released on exosomes. A high-content siRNA screen identified the endosomal sorting complexes required for transport (ESCRT)-associated protein ALIX as a regulator of both EGFR activity and PD-L1 surface presentation in basal-like breast cancer (BLBC) cells. ALIX depletion results in prolonged and enhanced stimulation-induced EGFR activity as well as defective PD-L1 trafficking through the MVB, reduced exosomal secretion, and its redistribution to the cell surface. Increased surface PD-L1 expression confers an EGFR-dependent immunosuppressive phenotype on ALIX-depleted cells. An inverse association between ALIX and PD-L1 expression was observed in human breast cancer tissues, while an immunocompetent mouse model of breast cancer revealed that ALIX-deficient tumors are larger and show an increased immunosuppressive environment. Our data suggest that ALIX modulates immunosuppression through regulation of PD-L1 and EGFR and may, therefore, present a diagnostic and therapeutic target for BLBC.

Tertiary lymphoid structures (TLS) and B cell infiltration are strong predictors of immunotherapy success across cancers, including triple-negative breast cancer (TNBC). However, immune-cold TNBCs often lack both features. Here, we identify a tumor-intrinsic mechanism that actively suppresses B cell recruitment. Despite evidence of B cell responses in cancer-associated lymph nodes (cLNs), B cells fail to infiltrate TNBC tumors or form TLS. This exclusion is not simply due to chemokine deficiency as exogenous chemokine addition fails to restore B cell migration. Using fractionation and metabolic profiling, we identify lactate as a dominant tumor-secreted metabolite that directly impairs B cell chemotaxis by disrupting mitochondrial metabolism. In vivo , combining lactate inhibition with engineered chemokine secretion promotes cLN-derived B cell infiltration and enables TLS formation, particularly when coupled with CD40 stimulation. Transcriptomics analyses across several human cancer datasets strengthen the association between high glycolytic activity with poor B-cell infiltration in chemokine-rich tumors. Together, our findings reveal lactate as a key metabolic barrier to B cell trafficking and TLS induction, suggesting that metabolic reprogramming may provide an avenue to convert “immune-cold” tumors into TLS-rich, immunologically responsive microenvironments. ### Competing Interest Statement A.G. CEO Pharos AI; D.P.C. is named inventor on a patent relating to synthetic lethality of NMT inhibitors in high-MYC cancers (WO2020128475); D.P.C. and M.S.H. are named inventors on a patent relating to Follicular Lymphoma biomarker signature (GB2509744.5). D.P.C. research funding from AstraZeneca. These competing interests are unrelated to this work. A.G. and D.P.C. Research funding - Boehringer Ingelheim. This competing interest is related to this work. All other authors declare no competing interests. Cancer Research UK, grant CC2078 Medical Research Council, https://ror.org/03x94j517, grant CC2078 Wellcome Trust, https://ror.org/029chgv08, grant CC2078 Medical Research Council, grant MR/W025221/1 Boehringer Ingelheim (Austria), grant OpnMe Biotechnology and Biological Sciences Research Council Biotechnology and Biological Sciences Research Council, BBS/E/B/000C0428 Cancer Research UK, grant CRUK/07/012 Breast Cancer Now, Toby Robins Research Centre at the ICR and Breast Cancer Now Unit at KCL CRUK City of London, CANTAC721\100023

Background Cancer research emphasises early detection, yet quantitative methods for analysing normal tissue remain limited. Hematoxylin and eosin (H&E)-stained tissues in digitised whole slide images (WSIs) enable computational histopathology; however, artificial intelligence (AI)-based analyses of normal breast tissue (NBT) remain scarce. Methods We curated 70 WSIs of NBTs from multiple sources with pathologist-guided manual annotations of epithelium, stroma, and adipocytes, and developed robust convolutional neural network (CNN)-based, patch-level classification models, named NBT-Classifiers , to tessellate and classify NBTs at different scales. Data and code are available at and . Findings Across three external cohorts, NBT-Classifiers trained on 128 x 128 µm and 256 x 256 µm patches achieved AUCs of 0·98–1·00. Two explainable artificial-intelligence (AI)- visualisation techniques confirmed the biological relevance of tissue class predictions. An end-to-end WSI pre-processing framework was then integrated, capable of localising lobules and peri-lobular stroma. The outputs are compatible with the QuPath v0.3.0 platform and enable downstream digital image analysis, such as texture and nuclei morphology assessment, at the patch level. Interpretation NBT-Classifiers represent a robust, generalisable, end-to-end deep learning framework, which will enable broader application in studies of normal tissues, in the context of breast and breast cancer. Funding The Breast Cancer Research Trust, Breast Cancer Now (and their legacy charity Breakthrough Breast Cancer), the Medical Research Council (MRC) [MR/X012476/1], Cancer Research UK [CRUK/07/012, KCL-BCN-Q3], and CRUK City of London Centre Award [CTRQQR-2021/100004]. Evidence before this study We searched PubMed from Jan 1, 2013, to Mar 20, 2025, using the query (("normal breast" OR "terminal duct lobular unit" OR "TDLU" OR "acini") AND ("artificial intelligence" OR "deep learning" OR "convolutional neural network" OR "CNN" OR "machine learning" OR "automation" OR "classification" OR "segmentation" OR "detection")) AND (journal article[pt] NOT review[pt]) NOT ("computed tomography" OR "CT" OR "ultrasound" OR "sonography" OR "US" OR "mammography" OR "mammogram" OR "Raman" OR "single-cell" OR "MRI"). This yielded 264 initial hits and after a thorough manual review, only eight studies were found to be relevant for using deep learning or machine learning to automate tissue recognition or quantification of normal breast tissues (NBTs) on digitised histological images. All these methods are highly specialised for detecting and quantifying lobules or quantifying tissue composition, and none of them are specifically tailored for providing patch-level tissue classifications on WSIs of NBTs. Moreover, NBTs are heavily underrepresented in large annotated WSI databases. A recent literature review summarised publicly available breast hematoxylin and eosin (H&E) WSI datasets from 2015 to 2023, identifying 17 datasets comprising a total of 10,385 breast H&E WSIs. Of these, two datasets contain female normal breast WSIs with pathology-guided manual annotations. Added value of this study Motivated by the research gap, we collected a comprehensive WSI dataset that captures real-world variability present across NBTs and provided by expert ground-truth manual annotations. Based on extensive cross-validation and external testing, we developed robust patch-level classification models to classify three major tissue compartments within NBTs, at different scales. Implications of all the available evidence With high AUCs of 0·98–1·00 across three external cohorts, our artificial-intelligence (AI) tool can be used to automatically classify tissue compartments and localise lobular regions for downstream normal breast research. These approaches have the potential to enhance our understanding of how various NBT components contribute to both benign and malignant breast pathology and lay the groundwork for the development of more advanced deep learning models and spatial-defined molecular large-scale analyses in the future. ### Competing Interest Statement Anita Grigoriadis, Louise J. Jones and Greg Verghese are Co-Founders of PharosAI. Salim Arslan and Pahini Pandya and employed by Panakeia Technology, UK. All other authors declare no conflict of interest. WSIs involved in this study are stored at the OASIS repository: , which currently can be accessed upon request. The Breast Cancer Research TrustThe Breast Cancer Research Trust, , Breast Cancer NowBreast Cancer Now, , the Medical Research Council (MRC)the Medical Research Council (MRC), , MR/X012476/1 Cancer Research UKCancer Research UK, , CRUK/07/012, KCL-BCN-Q3 CRUK City of London Centre AwardCRUK City of London Centre Award, , CTRQQR-2021/100004

INTRODUCTION:Toll-like receptor 9 (TLR9) is primarily expressed in human dendritic and B cells and recognizes double-stranded DNA motifs from pathogens to initiate an inflammatory response. Recent studies have revealed TLR9s' involvement beyond its conventional role in the immune response, notably during the tumorigenesis of various cancers such as head and neck, cervical, and ovarian cancers. METHODS:In this study patient biopsies of breast cancer tumors and normal breast epithelium were analyzed by immunohistochemistry to examine TLR9 expression. The study also investigated downregulation in transformed breast cancer cell lines compared to untransformed breast epithelial cells by analyzing gene or protein expression, including TLR9, IL-6, CCL2, CXCL1, and GM-CSF. MDA-MB-361 cells were engineered to express exogenous TLR9, and the effects on colony growth and senescence were assessed using colony formation assays, senescence staining, cytokine analysis, and flow cytometry. RESULTS:TLR9 levels in breast cancer tumors were significantly reduced compared to normal breast tissue epithelium. This downregulation was also observed in several transformed breast cancer cell lines compared to untransformed breast epithelial cell lines. Furthermore, MDA-MB-361 breast cancer cells expressing exogenous TLR9 exhibited reduced colony growth and an increase in the senescence marker IL-6, pro-inflammatory cytokine CCL2, CXCL1 chemokine; and growth factor GM-CSF. CONCLUSION:These findings support TLR9's regulatory role in mitigating breast cancer and highlight its critical connection between the innate immunity and tumor cell growth.

IgE antibodies directed against cancer antigens have demonstrated potent anti-tumour effects in pre-clinical studies. MOv18 IgE, the first-in-class IgE recognising the cancer antigen folate receptor alpha (FRα), showed preliminary signs of efficacy in a Phase I trial. Treatment was well tolerated, with the most common adverse event being transient urticarial skin reactions. We investigated immunological and allergic response parameters associated with urticarial skin reactions in MOv18 IgE-treated patients. Expression of target antigen, FRα, and MOv18 IgE reactivity with FRα or any component in human skin was studied by immunohistochemistry, immunofluorescence and immuno-mass spectrometry. We conducted transcriptomic analyses in paired lesional and non-lesional skin biopsies from a patient who developed an urticarial skin reaction. Systemic immunological markers including cytokines, β-tryptase and basophil activation states were interrogated throughout the trial and contemporaneously with the skin reaction. Of the 24 IgE-treated patients, 62.5% developed transient urticarial skin reactions, with onset during the first infusion, diminishing with consecutive infusions and no β-tryptase elevation nor clinical features indicating allergic aetiology. No FRα expression or MOv18 IgE binding to human skin was identified. Lesional skin biopsies from a patient given the highest antibody dose revealed scattered eosinophils, neutrophils and mast cell degranulation, but no increased immune cell infiltration. Transcriptomic analysis indicated pro-inflammatory, but not allergic, pathway activation. No systemic allergic or hypersensitivity mediators or basophil activation were detected. Urticarial skin reactions following MOv18 IgE treatment were unlikely to result from allergic mechanisms or skin antigen recognition. The clinical presentation is consistent with infusion-related reactions commonly observed with monoclonal antibody treatments. EudraCT number: 2014-000070-19; ClinicalTrials.gov identifier: NCT02546921, registered 11/Sept/2015.

Abstract Ageing, a breast cancer risk factor, can be reflected in histologically normal breast tissue (NBT). However, the relationship between age-related histological features and cancer risk is not fully understood. We propose that biological and microscopic features can be identified in seemingly histologically NBT in gBRCA1/2m carriers that are indicative of the earliest tissue changes of breast cancer, potentially as a result of accelerated tissue ageing. To study NBT, we have put in place a unique and scalable repository, named OASIS, storing currently >2,000 digitised whole slide images (WSI) of H&E-stained NBT from individuals with long clinical follow-up. These tissues are from individuals with a spectrum of risk for developing breast cancer and derived from reduction surgery from non-gBRCA1/2m carriers, risk reducing masectomies, and from contralateral and peri-tumoural NBT from gBRCA1/2m carriers and women with breast cancer. Across 6 NBT resources, 70 selected WSIs were manually annotated for epithelial cells, fibrous stroma and adipocytes, resembling the "ground truth". Extensive comparisons of different tile sizes, tile overlapping ratios, stain normalisation techniques and deep learning (DL) feature extractors were conducted to implement a Tissue-Classifier, which achieved 95% accuracy in 3-fold cross-validation. Based on the MobileNet architecture, a DL-framework achieved a sensitivity of 81.2% and specificity of 81.8% (84.7% AUC) in predicting Tissue-Age based on discernible histological patterns in healthy women. In gBRCA1/2m carriers, we discovered features of accelerated tissue ageing (e.g. degree of lobule involution), indicating that the biological NBT age differs from their chronological age. Our DL-based framework robustly captured histological patterns to predict tissue components and age. The global organisation of these patterns showed variations in WSIs of women with different risks of developing breast cancers, which may provide new insights into premalignant alterations in NBT. Citation Format: Anita Grigoriadis, Mario Parreno-Centeno, Siyuan Chen, Gregory Verghese, Esther Lips, Cheryl Gillett, Louise Jones, Sarah Pinder. Age-related normal breast tissue features differ in women with germline BRCA1/2 mutations [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PS03-09.

Checkpoint inhibition (CPI), particularly that targeting the inhibitory coreceptor programmed cell death protein 1 (PD-1), has transformed oncology. Although CPI can derepress cancer (neo)antigen-specific αβ T cells that ordinarily show PD-1-dependent exhaustion, it can also be efficacious against cancers evading αβ T cell recognition. In such settings, γδ T cells have been implicated, but the functional relevance of PD-1 expression by these cells is unclear. Here we demonstrate that intratumoral TRDV1 transcripts (encoding the TCRδ chain of Vδ1 + γδ T cells) predict anti-PD-1 CPI response in patients with melanoma, particularly those harboring below average neoantigens. Moreover, using a protocol yielding substantial numbers of tissue-derived Vδ1 + cells, we show that PD-1 + Vδ1 + cells display a transcriptomic program similar to, but distinct from, the canonical exhaustion program of colocated PD-1 + CD8 + αβ T cells. In particular, PD-1 + Vδ1 + cells retained effector responses to TCR signaling that were inhibitable by PD-1 engagement and derepressed by CPI.

PurposeThe Graham Roberts Study was initiated in 2018 and is the first Trials Within Cohorts (TwiCs) study for bladder cancer. Its purpose is to provide an infrastructure for answering a breadth of research questions, including clinical, mechanistic, and supportive care centred questions for bladder cancer patients.ParticipantsAll consented patients are those aged 18 or older, able to provide signed informedconsent and have a diagnosis of new or recurrent bladder cancer. All patients are required to have completed a series of baseline questionnaires. The questionnaires are then sent out every 12 months and include information on demographics and medical history as well as questionnaires to collect information on quality of life, fatigue, depression, overall health, physical activity, and dietary habits. Clinical information such as tumor stage, grade and treatment has also been extracted for each patient.Findings to dateTo date, a total of 125 bladder cancer patients have been consented onto the study with 106 filling in the baseline questionnaire. The cohort is made up of 75% newly diagnosed bladder cancer patients and 66% non-muscle invasive bladder cancer cases. At present, there is 1-year follow-up information for 70 patients, 2-year follow-up for 57 patients, 3-year follow-up for 47 patients and 4-year follow-up for 19 patients.Future plansWe plan to continue recruiting further patients into the cohort study. Using the data collected within the study, we hope to carry out independent research studies with a focus on quality of life. We are also committed to utilizing the Roberts Study Cohort to set up and commence an intervention. The future studies and trials carried out using the Roberts Cohort have the potential to identify and develop interventions that could improve the prevention, diagnosis, and treatment of bladder cancer.

AimsOver 50% of breast cancer cases are “Human epidermal growth factor receptor 2 (HER2) low breast cancer (BC)”, characterized by HER2 immunohistochemistry (IHC) scores of 1+ or 2+ alongside no amplification on fluorescence in situ hybridization (FISH) testing. The development of new anti‐HER2 antibody‐drug conjugates (ADCs) for treating HER2‐low breast cancers illustrates the importance of accurately assessing HER2 status, particularly HER2‐low breast cancer. In this study we evaluated the performance of a deep‐learning (DL) model for the assessment of HER2, including an assessment of the causes of discordances of HER2‐Null between a pathologist and the DL model. We specifically focussed on aligning the DL model rules with the ASCO/CAP guidelines, including stained cells' staining intensity and completeness of membrane staining.Methods and ResultsWe trained a DL model on a multicentric cohort of breast cancer cases with HER2‐IHC scores (n = 299). The model was validated on two independent multicentric validation cohorts (n = 369 and n = 92), with all cases reviewed by three senior breast pathologists. All cases underwent a thorough review by three senior breast pathologists, with the ground truth determined by a majority consensus on the final HER2 score among the pathologists. In total, 760 breast cancer cases were utilized throughout the training and validation phases of the study. The model's concordance with the ground truth (ICC = 0.77 [0.68–0.83]; Fisher P = 1.32e‐10) is higher than the average agreement among the three senior pathologists (ICC = 0.45 [0.17–0.65]; Fisher P = 2e‐3). In the two validation cohorts, the DL model identifies 95% [93% ‐ 98%] and 97% [91% ‐ 100%] of HER2‐low and HER2‐positive tumours, respectively. Discordant results were characterized by morphological features such as extended fibrosis, a high number of tumour‐infiltrating lymphocytes, and necrosis, whilst some artefacts such as nonspecific background cytoplasmic stain in the cytoplasm of tumour cells also cause discrepancy.ConclusionDeep learning can support pathologists' interpretation of difficult HER2‐low cases. Morphological variables and some specific artefacts can cause discrepant HER2‐scores between the pathologist and the DL model.

Abstract Introduction Human epidermal growth factor receptor 2 (HER2) protein overexpression and/or HER2 gene amplification is found in about 20% of invasive breast cancers. Considering the results of the DESTINY-Breast trials confirming the remarkable efficacy of anti-HER2 antibody drug-conjugate (T-Dxd) in both HER2-overexpressed and HER2-low tumors, it is necessary to identify not only HER2 (immunohistochemistry (IHC) score 3+) overexpressing tumors, but also HER2-low tumors. The latter category, defined as IHC1+ or IHC2+ but non-amplified, has proven to be challenging even for experienced pathologists, with high inter-observer variability. Here, we validate the performance of a deep learning (DL) model at: 1) predicting the IHC score from IHC histological features and 2) identifying HER2-low tumors with high sensitivity. Methods 675 HER2 stained IHC slides from primary breast cancer patients were selected based on pathology reports across three different cohorts (KCL_GSTT_1 n=369; Cypath_Breast n=214; KCL_GSTT_2 n=92). The slides were digitized and reviewed by expert breast pathologists. Specifically, Cypath_Breast and KCL_GSTT_2 were each annotated by one expert pathologist, while KCL_GSTT_1 was annotated by 5 expert pathologists with the ground truth defined as a majority vote between the 5 annotators. Cypath_Breast was assigned as the “discovery” set; while KCL_GSTT_1 and KCL_GSTT_2 were used as two independent validation cohorts. The model was specifically trained to extract features from IHC tiles to ensure they would capture both the staining intensity and the staining location on the cells (membrane, cytoplasmic or nuclei). The one-vs-one (OVO) and one-vs-rest (OVR) AUC as well as sensitivity and specificity to HER2-low and positive tumors were used as metrics to assess the performance of our model. An expert pathologist reviewed the most predictive regions of HER2 expression. Results After review by expert pathologists, 59 slides from KCL_GSTT_1 were removed either because of folded tissue, blurriness or because there was too little tissue. The IHC scores (0/1+/2+/3+) for the 3 cohorts were distributed as follows; Cypath_Breast: 42/120/36/16; KCL_GSTT_1: 120/90/52/48; KCL_GSTT_2: 54/22/15/1. For KCL_GSTT_1, the average agreement between the 5 expert pathologists amounted to a Cohen’s Kappa score of 0.63. Following training on Cypath_Breast, the performance of the model are presented in An ablation study proved that a feature extractor trained on IHC tiles outperformed an in-house extractor solely trained on H&E tiles (OVO AUC 0.95 vs 0.93 on KCL_GSTT_1; DeLong p< 0.001). Specifically, the model performed best at distinguishing HER2-negative (HER2 IHC 0) from HER2-low and HER2-positive tumors with a sensitivity (0 vs 1+/2+/3+) of 0.95 [0.93 - 0.98] and a specificity (0 vs 1+/2+/3+) of 0.78 [0.72 - 0.84] on KCL_GSTT_1. We observed the same tendency on KCL_GSTT_2 with a sensitivity (0 vs 1+/2+/3+) of 0.97 [0.91 - 1.00] and a specificity of 0.91 [0.83 - 0.98]. More importantly, the DL model reached a Cohen’s Kappa score of 0.72 [0.67 - 0.77] with the ground truth on KCL_GSTT_1, surpassing the average agreement Kappa score between expert pathologists (0.63). The most predictive regions showed cytoplasmic staining in tumor regions. Conclusion Our model provides a path towards a fully automated workflow for identifying up to 95% of breast cancer patients who could potentially benefit from anti-HER2 targeted therapies. Additionally, we show the utility of AI-based tools to minimize discrepancies among pathologists and assist in the diagnostic patient pathway. Table 1: Results of external validation on KCL_GSTT_1 and KCL_GSTT_2 Cohorts OVO AUC OVR AUC IHC 0 OVR AUC IHC 1+ OVR AUC IHC 2+ OVR AUC IHC 3+ KCL_GSTT_1 0.95 0.95 0.80 0.91 0.99 [0.93 - 0.96] [0.94 - 0.95] [0.79 - 0.82] [0.90 - 0.92] [0.99 - 1.00] KCL_GSTT_2 0.92 0.97 0.78 0.84 N/A [0.90 - 0.93] [0.96 - 0.98] [0.76 - 0.81] [0.80 - 0.88] Citation Format: Pierre-Antoine Bannier, Loïc Herpin, Rémy Dubois, Lydwine Van Praet, Charles Maussion, Ellen Amonoo, Anca Mera, Jasmine Timbres, Cheryl Gillett, Elinor Sawyer, Patrycja Gazinska, Piotr Ziolkowski, Roberto Salgado, Sheeba Irshad. Deep learning model for automated quantification of HER2 expression in invasive breast cancers from immunohistochemical whole slide images [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO2-07-05.

Cell type specific markers and genes

Supplementary Figures 1-7 from FGFR1 Amplification Drives Endocrine Therapy Resistance and Is a Therapeutic Target in Breast Cancer