Federated Learning for Decentralized Artificial Intelligence in Melanoma Diagnostics.

Importance: The development of artificial intelligence (AI)-based melanoma classifiers typically calls for large, centralized datasets, requiring hospitals to give away their patient data, which raises serious privacy concerns. To address this concern, decentralized federated learning has been proposed, where classifier development is distributed across hospitals. Objective: To investigate whether a more privacy-preserving federated learning approach can achieve comparable diagnostic performance to a classical centralized (ie, single-model) and ensemble learning approach for AI-based melanoma diagnostics. Design, Setting, and Participants: This multicentric, single-arm diagnostic study developed a federated model for melanoma-nevus classification using histopathological whole-slide images prospectively acquired at 6 German university hospitals between April 2021 and February 2023 and benchmarked it using both a holdout and an external test dataset. Data analysis was performed from February to April 2023. Exposures: All whole-slide images were retrospectively analyzed by an AI-based classifier without influencing routine clinical care. Main Outcomes and Measures: The area under the receiver operating characteristic curve (AUROC) served as the primary end point for evaluating the diagnostic performance. Secondary end points included balanced accuracy, sensitivity, and specificity. Results: The study included 1025 whole-slide images of clinically melanoma-suspicious skin lesions from 923 patients, consisting of 388 histopathologically confirmed invasive melanomas and 637 nevi. The median (range) age at diagnosis was 58 (18-95) years for the training set, 57 (18-93) years for the holdout test dataset, and 61 (18-95) years for the external test dataset; the median (range) Breslow thickness was 0.70 (0.10-34.00) mm, 0.70 (0.20-14.40) mm, and 0.80 (0.30-20.00) mm, respectively. The federated approach (0.8579; 95% CI, 0.7693-0.9299) performed significantly worse than the classical centralized approach (0.9024; 95% CI, 0.8379-0.9565) in terms of AUROC on a holdout test dataset (pairwise Wilcoxon signed-rank, P < .001) but performed significantly better (0.9126; 95% CI, 0.8810-0.9412) than the classical centralized approach (0.9045; 95% CI, 0.8701-0.9331) on an external test dataset (pairwise Wilcoxon signed-rank, P < .001). Notably, the federated approach performed significantly worse than the ensemble approach on both the holdout (0.8867; 95% CI, 0.8103-0.9481) and external test dataset (0.9227; 95% CI, 0.8941-0.9479). Conclusions and Relevance: The findings of this diagnostic study suggest that federated learning is a viable approach for the binary classification of invasive melanomas and nevi on a clinically representative distributed dataset. Federated learning can improve privacy protection in AI-based melanoma diagnostics while simultaneously promoting collaboration across institutions and countries. Moreover, it may have the potential to be extended to other image classification tasks in digital cancer histopathology and beyond.

Model soups improve performance of dermoscopic skin cancer classifiers.

BACKGROUND: Image-based cancer classifiers suffer from a variety of problems which negatively affect their performance. For example, variation in image brightness or different cameras can already suffice to diminish performance. Ensemble solutions, where multiple model predictions are combined into one, can improve these problems. However, ensembles are computationally intensive and less transparent to practitioners than single model solutions. Constructing model soups, by averaging the weights of multiple models into a single model, could circumvent these limitations while still improving performance. OBJECTIVE: To investigate the performance of model soups for a dermoscopic melanoma-nevus skin cancer classification task with respect to (1) generalisation to images from other clinics, (2) robustness against small image changes and (3) calibration such that the confidences correspond closely to the actual predictive uncertainties. METHODS: We construct model soups by fine-tuning pre-trained models on seven different image resolutions and subsequently averaging their weights. Performance is evaluated on a multi-source dataset including holdout and external components. RESULTS: We find that model soups improve generalisation and calibration on the external component while maintaining performance on the holdout component. For robustness, we observe performance improvements for pertubated test images, while the performance on corrupted test images remains on par. CONCLUSIONS: Overall, souping for skin cancer classifiers has a positive effect on generalisation, robustness and calibration. It is easy for practitioners to implement and by combining multiple models into a single model, complexity is reduced. This could be an important factor in achieving clinical applicability, as less complexity generally means more transparency.

Explainable artificial intelligence in skin cancer recognition: A systematic review.

BACKGROUND: Due to their ability to solve complex problems, deep neural networks (DNNs) are becoming increasingly popular in medical applications. However, decision-making by such algorithms is essentially a black-box process that renders it difficult for physicians to judge whether the decisions are reliable. The use of explainable artificial intelligence (XAI) is often suggested as a solution to this problem. We investigate how XAI is used for skin cancer detection: how is it used during the development of new DNNs? What kinds of visualisations are commonly used? Are there systematic evaluations of XAI with dermatologists or dermatopathologists? METHODS: Google Scholar, PubMed, IEEE Explore, Science Direct and Scopus were searched for peer-reviewed studies published between January 2017 and October 2021 applying XAI to dermatological images: the search terms histopathological image, whole-slide image, clinical image, dermoscopic image, skin, dermatology, explainable, interpretable and XAI were used in various combinations. Only studies concerned with skin cancer were included. RESULTS: 37 publications fulfilled our inclusion criteria. Most studies (19/37) simply applied existing XAI methods to their classifier to interpret its decision-making. Some studies (4/37) proposed new XAI methods or improved upon existing techniques. 14/37 studies addressed specific questions such as bias detection and impact of XAI on man-machine-interactions. However, only three of them evaluated the performance and confidence of humans using CAD systems with XAI. CONCLUSION: XAI is commonly applied during the development of DNNs for skin cancer detection. However, a systematic and rigorous evaluation of its usefulness in this scenario is lacking.

Skin cancer classification via convolutional neural networks: systematic review of studies involving human experts.

BACKGROUND: Multiple studies have compared the performance of artificial intelligence (AI)-based models for automated skin cancer classification to human experts, thus setting the cornerstone for a successful translation of AI-based tools into clinicopathological practice. OBJECTIVE: The objective of the study was to systematically analyse the current state of research on reader studies involving melanoma and to assess their potential clinical relevance by evaluating three main aspects: test set characteristics (holdout/out-of-distribution data set, composition), test setting (experimental/clinical, inclusion of metadata) and representativeness of participating clinicians. METHODS: PubMed, Medline and ScienceDirect were screened for peer-reviewed studies published between 2017 and 2021 and dealing with AI-based skin cancer classification involving melanoma. The search terms skin cancer classification, deep learning, convolutional neural network (CNN), melanoma (detection), digital biomarkers, histopathology and whole slide imaging were combined. Based on the search results, only studies that considered direct comparison of AI results with clinicians and had a diagnostic classification as their main objective were included. RESULTS: A total of 19 reader studies fulfilled the inclusion criteria. Of these, 11 CNN-based approaches addressed the classification of dermoscopic images; 6 concentrated on the classification of clinical images, whereas 2 dermatopathological studies utilised digitised histopathological whole slide images. CONCLUSIONS: All 19 included studies demonstrated superior or at least equivalent performance of CNN-based classifiers compared with clinicians. However, almost all studies were conducted in highly artificial settings based exclusively on single images of the suspicious lesions. Moreover, test sets mainly consisted of holdout images and did not represent the full range of patient populations and melanoma subtypes encountered in clinical practice.

A benchmark for neural network robustness in skin cancer classification.

BACKGROUND: One prominent application for deep learning-based classifiers is skin cancer classification on dermoscopic images. However, classifier evaluation is often limited to holdout data which can mask common shortcomings such as susceptibility to confounding factors. To increase clinical applicability, it is necessary to thoroughly evaluate such classifiers on out-of-distribution (OOD) data. OBJECTIVE: The objective of the study was to establish a dermoscopic skin cancer benchmark in which classifier robustness to OOD data can be measured. METHODS: Using a proprietary dermoscopic image database and a set of image transformations, we create an OOD robustness benchmark and evaluate the robustness of four different convolutional neural network (CNN) architectures on it. RESULTS: The benchmark contains three data sets-Skin Archive Munich (SAM), SAM-corrupted (SAM-C) and SAM-perturbed (SAM-P)-and is publicly available for download. To maintain the benchmark's OOD status, ground truth labels are not provided and test results should be sent to us for assessment. The SAM data set contains 319 unmodified and biopsy-verified dermoscopic melanoma (n = 194) and nevus (n = 125) images. SAM-C and SAM-P contain images from SAM which were artificially modified to test a classifier against low-quality inputs and to measure its prediction stability over small image changes, respectively. All four CNNs showed susceptibility to corruptions and perturbations. CONCLUSIONS: This benchmark provides three data sets which allow for OOD testing of binary skin cancer classifiers. Our classifier performance confirms the shortcomings of CNNs and provides a frame of reference. Altogether, this benchmark should facilitate a more thorough evaluation process and thereby enable the development of more robust skin cancer classifiers.

Combining CNN-based histologic whole slide image analysis and patient data to improve skin cancer classification.

BACKGROUND: Clinicians and pathologists traditionally use patient data in addition to clinical examination to support their diagnoses. OBJECTIVES: We investigated whether a combination of histologic whole slides image (WSI) analysis based on convolutional neural networks (CNNs) and commonly available patient data (age, sex and anatomical site of the lesion) in a binary melanoma/nevus classification task could increase the performance compared with CNNs alone. METHODS: We used 431 WSIs from two different laboratories and analysed the performance of classifiers that used the image or patient data individually or three common fusion techniques. Furthermore, we tested a naive combination of patient data and an image classifier: for cases interpreted as 'uncertain' (CNN output score <0.7), the decision of the CNN was replaced by the decision of the patient data classifier. RESULTS: The CNN on its own achieved the best performance (mean ± standard deviation of five individual runs) with AUROC of 92.30% ± 0.23% and balanced accuracy of 83.17% ± 0.38%. While the classification performance was not significantly improved in general by any of the tested fusions, naive strategy of replacing the image classifier with the patient data classifier on slides with low output scores improved balanced accuracy to 86.72% ± 0.36%. CONCLUSION: In most cases, the CNN on its own was so accurate that patient data integration did not provide any benefit. However, incorporating patient data for lesions that were classified by the CNN with low 'confidence' improved balanced accuracy.

Clinical and Genetic Tumor Characteristics of Responding and Non-Responding Patients to PD-1 Inhibition in Hepatocellular Carcinoma.

Immune checkpoint inhibitors (ICIs) belong to the therapeutic armamentarium in advanced hepatocellular carcinoma (HCC). However, only a minority of patients benefit from immunotherapy. Therefore, we aimed to identify indicators of therapy response. This multicenter analysis included 99 HCC patients. Progression-free (PFS) and overall survival (OS) were studied by Kaplan-Meier analyses for clinical parameters using weighted log-rank testing. Next-generation sequencing (NGS) was performed in a subset of 15 patients. The objective response (OR) rate was 19% median OS (mOS)16.7 months. Forty-one percent reached a PFS > 6 months; these patients had a significantly longer mOS (32.0 vs. 8.5 months). Child-Pugh (CP) A and B patients showed a mOS of 22.1 and 12.1 months, respectively. Ten of thirty CP-B patients reached PFS > 6 months, including 3 patients with an OR. Tumor mutational burden (TMB) could not predict responders. Of note, antibiotic treatment within 30 days around ICI initiation was associated with significantly shorter mOS (8.5 vs. 17.4 months). Taken together, this study shows favorable outcomes for OS with low AFP, OR, and PFS > 6 months. No specific genetic pattern, including TMB, could identify responders. Antibiotics around treatment initiation were associated with worse outcome, suggesting an influence of the host microbiome on therapy success.

Distinct Mutation Patterns Reveal Melanoma Subtypes and Influence Immunotherapy Response in Advanced Melanoma Patients.

The detection of somatic driver mutations by next-generation sequencing (NGS) is becoming increasingly important in the care of advanced melanoma patients. In our study, we evaluated the NGS results of 82 melanoma patients from clinical routine in 2017. Besides determining the tumor mutational burden (TMB) and annotation of all genetic driver alterations, we investigated their potential as a predictor for resistance to immune checkpoint inhibitors (ICI) and as a distinguishing feature between melanoma subtypes. Melanomas of unknown primary had a similar mutation pattern and TMB to cutaneous melanoma, which hints at its cutaneous origin. Besides the typical hotspot mutation in BRAF and NRAS, we frequently observed CDKN2A deletions. Acral and mucosal melanomas were dominated by CNV alterations affecting PDGFRA, KIT, CDK4, RICTOR, CCND2 and CHEK2. Uveal melanoma often had somatic SNVs in GNA11/Q and amplification of MYC in all cases. A significantly higher incidence of BRAF V600 mutations and EGFR amplifications, PTEN and TP53 deletions was found in patients with disease progression while on ICI. Thus, NGS might help to characterize melanoma subtypes more precisely and to identify possible resistance mechanisms to ICI therapy. Nevertheless, NGS based studies, including larger cohorts, are needed to support potential genetic ICI resistance mechanisms.

Dynamics of cell-free tumour DNA correlate with treatment response of head and neck cancer patients receiving radiochemotherapy.

PURPOSE: Definitive radiochemotherapy (RCTX) with curative intent is one of the standard treatment options in patients with locally advanced head and neck squamous cell carcinoma (HNSCC). Despite this intensive therapy protocol, disease recurrence remains an issue. Therefore, we tested the predictive capacity of liquid biopsies as a novel biomarker during RCTX in patients with HNSCC. MATERIAL AND METHODS: We sequenced the tumour samples of 20 patients with locally advanced HNSCC to identify driver mutations. Subsequently, we performed a longitudinal analysis of circulating tumour DNA (ctDNA) dynamics during RCTX. Deep sequencing and UMI-based error suppression for the identification of driver mutations and HPV levels in the plasma enabled treatment-response monitoring prior, during and after RCTX. RESULTS: In 85% of all patients ctDNA was detectable, showing a significant correlation with the gross tumour volume (p-value 0.032). Additionally, the tumour allele fraction in the plasma was negatively correlated with the course of treatment (p-value <0.05). If ctDNA was detectable at the first follow-up, disease recurrence was seen later on. Circulating HPV DNA (cvDNA) could be detected in three patients at high levels, showing a similar dynamic behaviour to the ctDNA throughout treatment, and disappeared after treatment. CONCLUSIONS: Monitoring RCTX treatment-response using liquid biopsy in patients with locally advanced HNSCC is feasible. CtDNA can be seen as a surrogate marker of disease burden, tightly correlating with the gross tumour volume prior to the treatment start. The observed kinetic of ctDNA and cvDNA showed a negative correlation with time and treatment dosage in most patients.

PET/MRI and genetic intrapatient heterogeneity in head and neck cancers.

PURPOSE: The relation between functional imaging and intrapatient genetic heterogeneity remains poorly understood. The aim of our study was to investigate spatial sampling and functional imaging by FDG-PET/MRI to describe intrapatient tumour heterogeneity. METHODS: Six patients with oropharyngeal cancer were included in this pilot study. Two tumour samples per patient were taken and sequenced by next-generation sequencing covering 327 genes relevant in head and neck cancer. Corresponding regions were delineated on pretherapeutic FDG-PET/MRI images to extract apparent diffusion coefficients and standardized uptake values. RESULTS: Samples were collected within the primary tumour (n = 3), within the primary tumour and the involved lymph node (n = 2) as well as within two independent primary tumours (n = 1). Genetic heterogeneity of the primary tumours was limited and most driver gene mutations were found ubiquitously. Slightly increasing heterogeneity was found between primary tumours and lymph node metastases. One private predicted driver mutation within a primary tumour and one in a lymph node were found. However, the two independent primary tumours did not show any shared mutations in spite of a clinically suspected field cancerosis. No conclusive correlation between genetic heterogeneity and heterogeneity of PET/MRI-derived parameters was observed. CONCLUSION: Our limited data suggest that single sampling might be sufficient in some patients with oropharyngeal cancer. However, few driver mutations might be missed and, if feasible, spatial sampling should be considered. In two independent primary tumours, both lesions should be sequenced. Our data with a limited number of patients do not support the concept that multiparametric PET/MRI features are useful to guide biopsies for genetic tumour characterization.

Radiogenomics in head and neck cancer: correlation of radiomic heterogeneity and somatic mutations in TP53, FAT1 and KMT2D.

PURPOSE: Genetic tumour profiles and radiomic features can be used to complement clinical information in head and neck squamous cell carcinoma (HNSCC) patients. Radiogenomics imply the potential to investigate complementarity or interrelations of radiomic and genomic features, and prognostic factors might be determined. The aim of our study was to explore radiogenomics in HNSCC. METHODS: For 20 HNSCC patients treated with primary radiochemotherapy, next-generation sequencing (NGS) of tumour and corresponding normal tissue was performed. In total, 327 genes were investigated by panel sequencing. Radiomic features were extracted from computed tomography data. A hypothesis-driven approach was used for radiogenomic correlations of selected image-based heterogeneity features and well-known driver gene mutations in HNSCC. RESULTS: The most frequently mutated driver genes in our cohort were TP53 (involved in cell cycle control), FAT1 (Wnt signalling, cell-cell contacts, migration) and KMT2D (chromatin modification). Radiomic features of heterogeneity did not correlate significantly with somatic mutations in TP53 or KMT2D. However, somatic mutations in FAT1 and smaller primary tumour volumes were associated with reduced radiomic intra-tumour heterogeneity. CONCLUSION: The landscape of somatic variants in our cohort is well in line with previous reports. An association of somatic mutations in FAT1 with reduced radiomic tumour heterogeneity could potentially elucidate the previously described favourable outcomes of these patients. Larger studies are needed to validate this exploratory data in the future.

Multi-omics discovery of exome-derived neoantigens in hepatocellular carcinoma.

BACKGROUND: Although mutated HLA ligands are considered ideal cancer-specific immunotherapy targets, evidence for their presentation is lacking in hepatocellular carcinomas (HCCs). Employing a unique multi-omics approach comprising a neoepitope identification pipeline, we assessed exome-derived mutations naturally presented as HLA class I ligands in HCCs. METHODS: In-depth multi-omics analyses included whole exome and transcriptome sequencing to define individual patient-specific search spaces of neoepitope candidates. Evidence for the natural presentation of mutated HLA ligands was investigated through an in silico pipeline integrating proteome and HLA ligandome profiling data. RESULTS: The approach was successfully validated in a state-of-the-art dataset from malignant melanoma, and despite multi-omics evidence for somatic mutations, mutated naturally presented HLA ligands remained elusive in HCCs. An analysis of extensive cancer datasets confirmed fundamental differences of tumor mutational burden in HCC and malignant melanoma, challenging the notion that exome-derived mutations contribute relevantly to the expectable neoepitope pool in malignancies with only few mutations. CONCLUSIONS: This study suggests that exome-derived mutated HLA ligands appear to be rarely presented in HCCs, inter alia resulting from a low mutational burden as compared to other malignancies such as malignant melanoma. Our results therefore demand widening the target scope for personalized immunotherapy beyond this limited range of mutated neoepitopes, particularly for malignancies with similar or lower mutational burden.

First case report of malignant peritoneal mesothelioma and oral verrucous carcinoma in a patient with a germline PTEN mutation: a combination of extremely rare diseases with probable further implications.

BACKGROUND: The PTEN-hamartoma-tumor-syndrome (PHTS) is caused by germline mutations in Phosphatase and Tensin homolog (PTEN) and predisposes to the development of several typical malignancies. Whereas PTEN mutations have been implicated in the occurrence of malignant mesotheliomas, the genetic landscape of verrucous carcinomas (VC) is largely uncharted. Both VC and malignant peritoneal mesotheliomas (MPM) are exceedingly rare and a potential link between these malignancies and PHTS has never been reported. CASE PRESENTATION: We here describe the clinical course of a PHTS patient who, in addition to a typical thyroid carcinoma at the age of 36 years, developed a highly-differentiated oral VC and an epithelioid MPM six years later. The patient with a history of occupational asbestos exposure underwent cytoreductive surgery and hyperthermic intraperitoneal chemotherapy for MPM. The clinical diagnosis of PHTS was consequently corroborated by a germline PTEN deletion. Sequencing of tumor tissue revealed a second hit in PTEN in the thyroid carcinoma and VC, confirmed by a PTEN loss and activation of the PI3K/AKT pathway in immunohistochemistry. Furthermore, additional somatic mutations in the thyroid carcinoma as well as in the VC were detected, whereas the genetics of MPM remained unrevealing. DISCUSSION AND CONCLUSIONS: We here report the very unusual clinical course of a patient with rare tumors that have a germline mutation first hit in PTEN in common. Since this patient was exposed to asbestos and current evidence suggests molecular mechanisms that might render PHTS patients particularly susceptible to mesothelioma, we strongly recommend PHTS patients to avoid even minimal exposure.

Circulating cell-free DNA: A potential biomarker to differentiate inflammation and infection during radiochemotherapy.

BACKGROUND AND PURPOSE: Radiochemotherapy is a standard treatment option for patients with head and neck cancer (HNSCC). During radiation, local toxicities are common and need to be differentiated from infections. As levels of circulating cell-free DNA (cfDNA) are known to be elevated during infections, cfDNA might complement clinical parameters. The aim of the study was to investigate the dynamics of cfDNA during radiochemotherapy. MATERIAL AND METHODS: In total, 78 blood samples of 20 patients with HNSCC were analysed in this prospective biomarker study. Blood samples were taken before and during treatment. CfDNA levels were quantified fluorometrically and results were compared to laboratory and clinical parameters. RESULTS: Elevated cfDNA levels were associated with the pre-treatment volumes of lymph node metastases (p = 0.0002), gastrostomy tube placement (20.23 ng/ml vs. 9.04 ng/ml (median), p = 0.025), the application of antibiotics (16.47 ng/ml vs. 9.04 ng/ml, p = 0.006) and manifest infections (16.81 ng/ml vs. 9.04 ng/ml, p = 0.010). Furthermore, a significant difference between moderate inflammation (radiation-induced toxicity RTOG grade 2-3) and manifest infections could be observed (8.97 ng/ml vs. 16.81 ng/ml, p = 0.014), allowing for a more pronounced differentiation than by CRP levels (p = 0.119). There might be an association between the application of G-CSF and elevated cfDNA levels. CONCLUSION: CfDNA levels are correlated with infections during radiochemotherapy and could represent an informative complemental biomarker to drive therapeutic decision-making. Estimated levels of circulating cell-free tumour DNA (ctDNA) in plasma should be interpreted cautiously when monitoring tumour outcome by next-generation-sequencing, as confounders like infections or drug application might influence the fraction of ctDNA in total cfDNA.

A comprehensive quality control workflow for paired tumor-normal NGS experiments.

SUMMARY: Quality control (QC) is an important part of all NGS data analysis stages. Many available tools calculate QC metrics from different analysis steps of single sample experiments (raw reads, mapped reads and variant lists). Multi-sample experiments, as sequencing of tumor-normal pairs, require additional QC metrics to ensure validity of results. These multi-sample QC metrics still lack standardization. We therefore suggest a new workflow for QC of DNA sequencing of tumor-normal pairs. With this workflow well-known single-sample QC metrics and additional metrics specific for tumor-normal pairs can be calculated. The segmentation into different tools offers a high flexibility and allows reuse for other purposes. All tools produce qcML, a generic XML format for QC of -omics experiments. qcML uses quality metrics defined in an ontology, which was adapted for NGS. AVAILABILITY AND IMPLEMENTATION: All QC tools are implemented in C ++ and run both under Linux and Windows. Plotting requires python 2.7 and matplotlib. The software is available under the 'GNU General Public License version 2' as part of the ngs-bits project: https://github.com/imgag/ngs-bits. CONTACT: christopher.schroeder@med.uni-tuebingen.de. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Personalized peptide vaccine-induced immune response associated with long-term survival of a metastatic cholangiocarcinoma patient.

BACKGROUND & AIMS: We report a novel experimental immunotherapeutic approach in a patient with metastatic intrahepatic cholangiocarcinoma. In the 5year course of the disease, the initial tumor mass, two local recurrences and a lung metastasis were surgically removed. Lacking alternative treatment options, aiming at the induction of anti-tumor T cells responses, we initiated a personalized multi-peptide vaccination, based on in-depth analysis of tumor antigens (immunopeptidome) and sequencing. METHODS: Tumors were characterized by immunohistochemistry, next-generation sequencing and mass spectrometry of HLA ligands. RESULTS: Although several tumor-specific neo-epitopes were predicted in silico, none could be validated by mass spectrometry. Instead, a personalized multi-peptide vaccine containing non-mutated tumor-associated epitopes was designed and applied. Immunomonitoring showed vaccine-induced T cell responses to three out of seven peptides administered. The pulmonary metastasis resected after start of vaccination showed strong immune cell infiltration and perforin positivity, in contrast to the previous lesions. The patient remains clinically healthy, without any radiologically detectable tumors since March 2013 and the vaccination is continued. CONCLUSIONS: This remarkable clinical course encourages formal clinical studies on adjuvant personalized peptide vaccination in cholangiocarcinoma. LAY SUMMARY: Metastatic cholangiocarcinomas, cancers that originate from the liver bile ducts, have very limited treatment options and a fatal prognosis. We describe a novel therapeutic approach in such a patient using a personalized multi-peptide vaccine. This vaccine, developed based on the characterization of the patient's tumor, evoked detectable anti-tumor immune responses, associating with long-term tumor-free survival.