Patients with head‐and‐neck cancer can develop both lung metastasis and primary lung cancer during the course of their disease. Despite the clinical importance of discrimination, reliable diagnostic biomarkers are still lacking. Here, we have characterised a cohort of squamous cell lung (SQCLC) and head‐and‐neck (HNSCC) carcinomas by quantitative proteomics. In a training cohort, we quantified 4,957 proteins in 44 SQCLC and 30 HNSCC tumours. A total of 518 proteins were found to be differentially expressed between SQCLC and HNSCC, and some of these were identified as genetic dependencies in either of the two tumour types. Using supervised machine learning, we inferred a proteomic signature for the classification of squamous cell carcinomas as either SQCLC or HNSCC, with diagnostic accuracies of 90.5% and 86.8% in cross‐ and independent validations, respectively. Furthermore, application of this signature to a cohort of pulmonary squamous cell carcinomas of unknown origin leads to a significant prognostic separation. This study not only provides a diagnostic proteomic signature for classification of secondary lung tumours in HNSCC patients, but also represents a proteomic resource for HNSCC and SQCLC.

Introduction: Management of esophageal anastomotic leaks (AL) and esophageal perforations (EP) remains difficult and often requires an interdisciplinary treatment modality. For primary endoscopic management, self-expanding metallic stent (SEMS) placement is often considered first-line therapy. Recently, endoscopic vacuum therapy (EVT) has emerged as an alternative or adjunct for management of these conditions. So far, data for EVT in the upper gastrointestinal-tract is restricted to single centre, non-randomized trials. No studies on optimal negative pressure application during EVT exist. The aim of our study is to describe our centre’s experience with low negative pressure (LNP) EVT for these indications over the past 5-years. Patients and Methods: Between January 2014 and December 2018, 30 patients were endoscopically treated for AL (n = 23) or EP (n = 7). All patients were primarily treated with EVT and LNP between –20 and –50 mm Hg. Additional endoscopic treatment was added when EVT failed. Procedural and peri-procedural data, as well as clinical outcomes including morbidity and mortality, were analysed. Results: Clinical successful endoscopic treatment of EP and AL was achieved in 83.3% (n = 25/30), with 73.3% success using EVT alone (n = 22/30). Mean treatment duration until leak closure was 16.1 days (range 2–58 days). Additional treatment modalities for complete leak resolution was necessary in 10% (n = 3/30), including SEMS placement and fibrin glue injection. Mean hospital stay for patients with EP was shorter with 33.7 days compared to AL with 54.4 days (p = 0.08). Estimated preoperative 10-year overall survival (Charlson comorbidity score) was 39.4% in patients with AL and 59.9% in patients with EP (p = 0.26). A mean of 5.1 EVT changes (range 1–12) was needed in EP and 3.6 changes (range 1–13) in AL to achieve complete closure, switch to other treatment modality, or reach endoscopic failure (p = 0.38). Conclusion: LNP EVT enables effective minimally – invasive endoluminal leak closure from anastomotic esophageal leaks and EP in high-morbid patients. In this study, EVT was combined with other endoscopic treatment options such as SEMS placement or fibrin glue injection in order to achieve leak or perforation closure in the vast majority of patients (83.3%). Low aspiration pressures led to slower but still sufficient clinical results.

The aim of this study was to discriminate malignant and benign clinical T1 renal masses on routinely acquired computed tomography (CT) images using radiomics and machine learning techniques.Adult patients undergoing surgical resection and histopathological analysis of clinical T1 renal masses were included. Preoperative CT studies in venous phase from multiple referring centers were included, without restriction to specific CT scanners, slice thickness, or degrees of artifacts. Renal masses were segmented and 120 standardized radiomic features extracted. Machine learning algorithms were used to predict malignancy of renal masses using radiomics features and cross-validation. Diagnostic accuracy of machine learning models and assessment by independent blinded radiologists were compared based on the gold standard of histopathologic diagnosis.A total of 94 patients met inclusion criteria (benign renal masses: n = 18; malignant: n = 76). CT studies from 18 different scanners were assessed with median slice thickness of 2.5 mm and artifacts in 15 cases (15.9%).Area under the receiver-operating-characteristics curve (AUC) of random forest (random forest [RF], AUC = 0.83) was significantly higher compared to the radiologists (AUC = 0.68, P = .047). Sensitivity was significantly higher for RF versus radiologists (0.88 vs 0.80, P = .045), whereas specificity was numerically higher for RF (0.67 vs 0.50, P = .083).Although limited by an overall small sample size and few benign renal tumors, a radiomic features and machine learning approach suggests a high diagnostic accuracy for discrimination of malignant and benign clinical T1 renal masses on venous phase CT. The presented algorithm robustly outperforms human readers in a real-life scenario with nonstandardized imaging studies from various referring centers. Abbreviations: AML = angiomyolipoma, AUC = area under the ROC curve, CD = cluster of differentiation, CK = cytokeratin, CT = computed tomography, HE = hematoxylin-eosin, HMB = human melanoma black, ICC = interobserver correlation coefficient, IQR = interquartile range, KNN = k-nearest neighbor, NN = neural network, POM = probability of malignancy, RCC = renal cell carcinoma, RF = random forest, RFE = recursive feature elimination, ROC = receiver operating characteristics, ROI = region of interest, SVM = c, US = United States, XG boost = extreme gradient boosting.

No abstract

No abstract