Background A methylation-based classification of ependymoma has recently found broad application. However, the diagnostic advantage and implications for treatment decisions remain unclear. Here, we retrospectively evaluate the impact of surgery and radiotherapy on outcome after molecular reclassification of adult intracranial ependymomas. Methods Tumors diagnosed as intracranial ependymomas from 170 adult patients collected from 8 diagnostic institutions were subjected to DNA methylation profiling. Molecular classes, patient characteristics, and treatment were correlated with progression-free survival (PFS). Results The classifier indicated an ependymal tumor in 73.5%, a different tumor entity in 10.6%, and non-classifiable tumors in 15.9% of cases, respectively. The most prevalent molecular classes were posterior fossa ependymoma group B (EPN-PFB, 32.9%), posterior fossa subependymoma (PF-SE, 25.9%), and supratentorial ZFTA fusion-positive ependymoma (EPN-ZFTA, 11.2%). With a median follow-up of 60.0 months, the 5- and 10-year-PFS rates were 64.5% and 41.8% for EPN-PFB, 67.4% and 45.2% for PF-SE, and 60.3% and 60.3% for EPN-ZFTA. In EPN-PFB, but not in other molecular classes, gross total resection (GTR) (P = .009) and postoperative radiotherapy (P = .007) were significantly associated with improved PFS in multivariable analysis. Histological tumor grading (WHO 2 vs. 3) was not a predictor of the prognosis within molecularly defined ependymoma classes. Conclusions DNA methylation profiling improves diagnostic accuracy and risk stratification in adult intracranial ependymoma. The molecular class of PF-SE is unexpectedly prevalent among adult tumors with ependymoma histology and relapsed as frequently as EPN-PFB, despite the supposed benign nature. GTR and radiotherapy may represent key factors in determining the outcome of EPN-PFB patients.

BACKGROUND:The COVID-19 pandemic has greatly increased the incidence and clinical importance of critical illness myopathy (CIM), because it is one of the most common complications of modern intensive care medicine. Current diagnostic criteria only allow diagnosis of CIM at an advanced stage, so that patients are at risk of being overlooked, especially in early stages. To determine the frequency of CIM and to assess a recently proposed tool for early diagnosis, we have followed a cohort of COVID-19 patients with acute respiratory distress syndrome and compared the time course of muscle excitability measurements with the definite diagnosis of CIM.METHODS:Adult COVID-19 patients admitted to the Intensive Care Unit of the University Hospital Bern, Switzerland requiring mechanical ventilation were recruited and examined on Days 1, 2, 5, and 10 post-intubation. Clinical examination, muscle excitability measurements, medication record, and laboratory analyses were performed on all study visits, and additionally nerve conduction studies, electromyography and muscle biopsy on Day 10. Muscle excitability data were compared with a cohort of 31 age-matched healthy subjects. Diagnosis of definite CIM was made according to the current guidelines and was based on patient history, results of clinical and electrophysiological examinations as well as muscle biopsy.RESULTS:Complete data were available in 31 out of 44 recruited patients (mean [SD] age, 62.4 [9.8] years). Of these, 17 (55%) developed CIM. Muscle excitability measurements on Day 10 discriminated between patients who developed CIM and those who did not, with a diagnostic precision of 90% (AUC 0.908; 95% CI 0.799-1.000; sensitivity 1.000; specificity 0.714). On Days 1 and 2, muscle excitability parameters also discriminated between the two groups with 73% (AUC 0.734; 95% CI 0.550-0.919; sensitivity 0.562; specificity 0.857) and 82% (AUC 0.820; CI 0.652-0.903; sensitivity 0.750; specificity 0.923) diagnostic precision, respectively. All critically ill COVID-19 patients showed signs of muscle membrane depolarization compared with healthy subjects, but in patients who developed CIM muscle membrane depolarization on Days 1, 2 and 10 was more pronounced than in patients who did not develop CIM.CONCLUSIONS:This study reports a 55% prevalence of definite CIM in critically ill COVID-19 patients. Furthermore, the results confirm that muscle excitability measurements may serve as an alternative method for CIM diagnosis and support its use as a tool for early diagnosis and monitoring the development of CIM.

Machine learning-based solutions rely heavily on the quality and quantity of the training data. In the medical domain, the main challenge is to acquire rich and diverse annotated datasets for training. We propose to decrease the annotation efforts and further diversify the dataset by introducing an annotation-efficient learning workflow. Instead of costly pixel-level annotation, we require only image-level labels as the remainder is covered by simulation. Thus, we obtain a large-scale dataset with realistic images and accurate ground truth annotations. We use this dataset for the instrument localization activity task together with a studentteacher approach. We demonstrate the benefits of our workflow compared to state-of-the-art methods in instrument localization that are trained only on clinical datasets, which are fully annotated by human experts.

Due to the limited amount of available annotated data in the medical field, domain generalization for applications in computer-assisted surgery is essential. Our work addresses this problem for the task of surgical instrument tip localization in neurosurgery, which is a classical step towards computer-assisted surgery. We propose an uncertainty-based CNN approach that dynamically selects the most relevant data source by incorporating its own uncertainty into the inference. In addition, the estimated uncertainty can visualize and easily explain the network’s decision. Quantitative and qualitative evaluations show that our method outperforms state of the art approaches for large domain shifts and results are on-par for in-domain applications. Further increasing domain shifts by testing on different surgical disciplines, eye and laparoscopic surgeries, proves the generalization capabilities of the proposed method.

Towards computer-assisted neurosurgery, scene understanding algorithms for microscope video data are required. Previous work utilizes optical flow to extract spatiotemporal context from neurosurgical video sequences. However, to select an appropriate optical flow method, we need to analyze which algorithm yields the highest accuracy for the neurosurgical domain. Currently, there are no benchmark datasets available for neurosurgery. In our work, we present an approach to generate synthetic data for optical flow evaluation on the neurosurgical domain. We simulate image sequences and thereby take into account domainspecific visual conditions such as surgical instrument motion. Then, we evaluate two optical flow algorithms, Farneback and PWC-Net, on our synthetic data. Qualitative and quantitative assessments confirm that our data can be used to evaluate optical flow for the neurosurgical domain. Future work will concentrate on extending the method by modeling additional effects in neurosurgery such as elastic background motion.

Towards computer-assisted neurosurgery, robust methods for instrument localization on neurosurgical microscope video data are needed. Specifically for neurosurgical data, challenges arise from visual conditions such as strong blur and from an unknowingly large variety of instrument types. For neurosurgical domain, instrument localization methods must generalize across different sub-disciplines such as cranial tumor and aneurysm surgeries which exhibit different visual properties. We present and evaluate a methodology towards robust instrument tip localization for neurosurgical microscope data, formulated as coarse saliency prediction. For our analysis, we build a comprehensive dataset comprising in-the-wild data from several neurosurgical sub-disciplines as well as phantom surgeries. Comparing single stream networks using either image or optical flow information, we find complementary performance of both networks. Plain optical flow enables better cross-domain generalization, while the image-based network performs better on surgeries from the training domain. Based on these findings, we present a two-stream architecture that fuses image and optical flow information to utilize the complementary performance of both. Being trained on tumor surgeries, our architecture outperforms both single stream networks and shows improved robustness on data from different neurosurgical sub-disciplines. From our findings, future work must focus more on how to incorporate optical flow information into fusion architectures to further improve cross-domain generalization.

Among the most challenging diagnostic issues in surgical neuropathology is the distinction between scant infiltration by diffuse gliomas and reactive gliosis. The best documented ancillary marker to establish a definitive diagnosis of glioma in this setting is the identification of hotspot mutations in the isocitrate dehydrogenase 1 and 2 (IDH1/IDH2) genes, which is limited, however, by the low prevalence of these mutations in gliomas of elderly adults. Since telomerase reverse transcriptase (TERT) promoter mutations are present in the vast majority of IDH-wildtype diffuse gliomas, we hypothesized that combined analysis of IDH and TERT might overcome these limitations. For this purpose, we analyzed a series of non-neoplastic and neoplastic CNS samples for the prevalence of TERT hotspot mutations. TERT mutations were identified in none out of 58 (0%) reactive gliosis samples, and in 91 out of 117 (78%) IDH-wildtype gliomas. Based on a series of 200 consecutive diffuse gliomas, we found that IDH mutation analysis alone had a sensitivity of 28% (63% and 12%, respectively, in patients below and above age of 50) for detection of gliomas, whereas a combined analysis of IDH and TERT was 85% sensitive (87% and 84%, respectively, below and above age of 50). In sum, our findings suggest that TERT promoter mutation analysis contributes favorably to a molecular panel in cases equivocal for glioma versus gliosis on morphological grounds, especially in patients above age of 50, in which IDH analysis alone performs poorly.

BACKGROUND:In contrast to isocitrate dehydrogenase (IDH) mutation analysis, which is homogenous within a given tumor, diagnostic errors in histological analysis following the 2016 World Health Organization (WHO) classification could be due to small samples because of histological heterogeneity. OBJECTIVE:To assess whether the sample size sent to histopathology influences the tumor grading in IDH wild-type gliomas. METHODS:Histologically diagnosed WHO grade, sample volume, and preoperative tumor volume data of 111 patients aged who received resection of IDHwt gliomas between January 2007 and December 2015 at our hospital were evaluated. The differences between absolute and relative pathological sample sizes stratified by WHO grade were conducted using One-Way-Permutation-Test. RESULTS:With a mean sample size of 10.9 cc, 83.8% of patients were histologically diagnosed as WHO grade IV, while 16.2% of patients with a mean sample size of 2.62 cc were diagnosed as WHO grade II/III. One-Way-Permutation-Test showed a significant difference between absolute tissue samples stratified by WHO grade (P = .0374). The distribution of preoperative tumor volumes with WHO grade IV vs WHO grade II/III showed no significant difference (P = .8587). Of all tumors with a sample size >10 cc 100% were pathologically diagnosed as WHO grade IV and those with sample size >5 cc 93.5% were diagnosed as WHO grade IV. CONCLUSION:Small sample sizes are associated with a higher risk of under-estimating malignancy in histological grading in IDHwt gliomas. This study suggests a standard minimum sample size (>5cc) in every resection. Modalities of adjuvant treatment for IDHwt, WHO grade II/III gliomas need to reflect a prognosis that is only marginally better than of a glioblastoma.

IDH (isocitrate dehydrogenase) gene mutations are present in most diffuse low-grade gliomas and define the clinico-pathological core of the respective morphologically defined entities. Conversely, according to the 2016 WHO classification, the majority of glioblastomas belong to the IDH-wildtype category, which is defined by exclusion. TERT (telomerase reverse transcriptase gene) promoter mutations have been suggested as a molecular marker for primary glioblastomas. We analyzed molecular, histopathological, and clinical profiles of a series of 110 consecutive diffuse gliomas (WHO grades II-IV) diagnosed at our institution, in which TERT promoter mutation analysis had been performed as part of diagnostic work-up. A diagnostic algorithm based on IDH, TERT, ATRX, H3F3A, and 1p19q co-deletion status resulted in a consistent molecular classification with only 14 (13%) marker-negative tumors. TERT promoter mutations were present in 77% of IDH-wildtype tumors. The TERT/IDH-wildtype category was highly enriched for tumors with unconventional clinical or histological features. Molecular classes were associated with distinct rates of MGMT promoter methylation. We conclude that, in a routine diagnostic setting, TERT promoter mutations define a relatively homogeneous core group among IDH-wildtype diffuse gliomas that includes the majority of primary glioblastomas as well as their putative precursor lesions.