Deep Learning Versus Neurologists: Functional Outcome Prediction in LVO Stroke Patients Undergoing Mechanical Thrombectomy

BACKGROUND

Despite evolving treatments, functional recovery in patients with large vessel occlusion stroke remains variable and outcome prediction challenging. Can we improve estimation of functional outcome with interpretable deep learning models using clinical and magnetic resonance imaging data?

METHODS

In this observational study, we collected data of 222 patients with middle cerebral artery M1 segment occlusion who received mechanical thrombectomy. In a 5-fold cross validation, we evaluated interpretable deep learning models for predicting functional outcome in terms of modified Rankin scale at 3 months using clinical variables, diffusion weighted imaging and perfusion weighted imaging, and a combination thereof. Based on 50 test patients, we compared model performances to those of 5 experienced stroke neurologists. Prediction performance for ordinal (modified Rankin scale score, 0-6) and binary (modified Rankin scale score, 0-2 versus 3-6) functional outcome was assessed using discrimination and calibration measures like area under the receiver operating characteristic curve and accuracy (percentage of correctly classified patients).

RESULTS

In the cross validation, the model based on clinical variables and diffusion weighted imaging achieved the highest binary prediction performance (area under the receiver operating characteristic curve, 0.766 [0.727-0.803]). Performance of models using clinical variables or diffusion weighted imaging only was lower. Adding perfusion weighted imaging did not improve outcome prediction. On the test set of 50 patients, binary prediction performance between model (accuracy, 60% [55.4%-64.4%]) and neurologists (accuracy, 60% [55.8%-64.21%]) was similar when using clinical data. However, models significantly outperformed neurologists when imaging data were provided, alone or in combination with clinical variables (accuracy, 72% [67.8%-76%] versus 64% [59.8%-68.4%] with clinical and imaging data). Prediction performance of neurologists with comparable experience varied strongly.

CONCLUSIONS

We hypothesize that early prediction of functional outcome in large vessel occlusion stroke patients may be significantly improved if neurologists are supported by interpretable deep learning models.

Deep transformation models for functional outcome prediction after acute ischemic stroke

In many medical applications, interpretable models with high prediction performance are sought. Often, those models are required to handle semistructured data like tabular and image data. We show how to apply deep transformation models (DTMs) for distributional regression that fulfill these requirements. DTMs allow the data analyst to specify (deep) neural networks for different input modalities making them applicable to various research questions. Like statistical models, DTMs can provide interpretable effect estimates while achieving the state-of-the-art prediction performance of deep neural networks. In addition, the construction of ensembles of DTMs that retain model structure and interpretability allows quantifying epistemic and aleatoric uncertainty. In this study, we compare several DTMs, including baseline-adjusted models, trained on a semistructured data set of 407 stroke patients with the aim to predict ordinal functional outcome three months after stroke. We follow statistical principles of model-building to achieve an adequate trade-off between interpretability and flexibility while assessing the relative importance of the involved data modalities. We evaluate the models for an ordinal and dichotomized version of the outcome as used in clinical practice. We show that both tabular clinical and brain imaging data are useful for functional outcome prediction, whereas models based on tabular data only outperform those based on imaging data only. There is no substantial evidence for improved prediction when combining both data modalities. Overall, we highlight that DTMs provide a powerful, interpretable approach to analyzing semistructured data and that they have the potential to support clinical decision-making.

Distributional anchor regression

Prediction models often fail if train and test data do not stem from the same distribution. Out-of-distribution (OOD) generalization to unseen, perturbed test data is a desirable but difficult-to-achieve property for prediction models and in general requires strong assumptions on the data generating process (DGP). In a causally inspired perspective on OOD generalization, the test data arise from a specific class of interventions on exogenous random variables of the DGP, called anchors. Anchor regression models, introduced by Rothenhäusler et al. (J R Stat Soc Ser B 83(2):215-246, 2021. 10.1111/rssb.12398), protect against distributional shifts in the test data by employing causal regularization. However, so far anchor regression has only been used with a squared-error loss which is inapplicable to common responses such as censored continuous or ordinal data. Here, we propose a distributional version of anchor regression which generalizes the method to potentially censored responses with at least an ordered sample space. To this end, we combine a flexible class of parametric transformation models for distributional regression with an appropriate causal regularizer under a more general notion of residuals. In an exemplary application and several simulation scenarios we demonstrate the extent to which OOD generalization is possible.

Machine-learning-based outcome prediction in stroke patients with middle cerebral artery-M1 occlusions and early thrombectomy

Background and purpose

Clinical outcomes vary substantially among individuals with large vessel occlusion (LVO) stroke. A small infarct core and large imaging mismatch were found to be associated with good recovery. The aim of this study was to investigate whether those imaging variables would improve individual prediction of functional outcome after early (<6 h) endovascular treatment (EVT) in LVO stroke.

Methods

We included 222 patients with acute ischemic stroke due to middle cerebral artery (MCA)‐M1 occlusion who received EVT. As predictors, we used clinical variables and region of interest (ROI)‐based magnetic resonance imaging features. We developed different machine‐learning models and quantified their prediction performance according to the area under the receiver‐operating characteristic curves and the Brier score.

Results

The rate of successful recanalization was 78%, with 54% patients having a favorable outcome (modified Rankin scale score 0–2). Small infarct core was associated with favorable functional outcome. Outcome prediction improved only slightly when imaging was added to patient variables. Age was the driving factor, with a sharp decrease in likelihood of favorable functional outcome above the age of 78 years.

Conclusions

In patients with MCA‐M1 occlusion strokes referred to EVT within 6 h of symptom onset, infarct core volume was associated with outcome. However, ROI‐based imaging variables led to no significant improvement in outcome prediction at an individual patient level when added to a set of clinical predictors. Our study is in concordance with current practice, where imaging mismatch or collateral readouts are not recommended as factors for excluding patients with MCA‐M1 occlusion for early EVT.

Neurodevelopmental Outcomes at Age 5 Years After Prophylactic Early High-Dose Recombinant Human Erythropoietin for Neuroprotection in Very Preterm Infants

This study reports 5-year neurodevelopmental outcomes for Swiss children born before 32 weeks’ gestation and randomized at birth to receive early high-dose recombinant human erythropoietin (rhEpo) vs placebo.

A prospective observational study on trajectories and prognostic factors of mid back pain

Background

Although mid back pain (MBP) is a common condition that causes significant disability, it has received little attention in research and knowledge about trajectories and prognosis of MBP is limited. The purpose of this study was to identify trajectories of MBP and baseline risk factors for an unfavorable outcome in MBP patients undergoing chiropractic treatment.

Methods

This prospective-observational study analyzes outcome data of 90 adult MBP patients (mean age = 37.0 ± 14.6 years; 49 females) during one year (at baseline, after 1 week, 1 month, 3, 6 and 12 months) after start of chiropractic treatment. Patients completed an 11-point (0 to 10) numeric pain rating scale (NRS) at baseline and one week, one month, three, six and twelve months after treatment start and the Patient’s Global Impression of Change (PGIC) questionnaire at all time points except baseline. To determine trajectories, clustering with the package kml (software R), a variant of k-means clustering adapted for longitudinal data, was performed using the NRS-data. The identified NRS-clusters and PGIC data after three months were tested for association with baseline variables using univariable logistic regression analyses, conditional inference trees and random forest plots.

Results

Two distinct NRS-clusters indicating a favourable (rapid improvement within one month from moderate pain to persistent minor pain or recovery after one year, 80% of patients) and an unfavourable trajectory (persistent moderate to severe pain, 20% of patients) were identified. Chronic (> 3 months) pain duration at baseline significantly predicted that a patient was less likely to follow a favourable trajectory [OR = 0.16, 95% CI = 0.05–0.50, p = 0.002] and to report subjective improvement after twelve months [OR = 0.19, 95% CI = 0.07–0.51, p = 0.001], which was confirmed by the conditional inference tree and the random forest analyses.

Conclusions

This prospective exploratory study identified two distinct MBP trajectories, representing a favourable and an unfavourable outcome over the course of one year after chiropractic treatment. Pain chronicity was the factor that influenced outcome measures using NRS or PGIC.

Developing Deep Learning Applications for Life Science and Pharma Industry

Deep Learning has boosted artificial intelligence over the past 5 years and is seen now as one of the major technological innovation areas, predicted to replace lots of repetitive, but complex tasks of human labor within the next decade. It is also expected to be 'game changing' for research activities in pharma and life sciences, where large sets of similar yet complex data samples are systematically analyzed. Deep learning is currently conquering formerly expert domains especially in areas requiring perception, previously not amenable to standard machine learning. A typical example is the automated analysis of images which are typically produced en-masse in many domains, e. g., in high-content screening or digital pathology. Deep learning enables to create competitive applications in so-far defined core domains of 'human intelligence'. Applications of artificial intelligence have been enabled in recent years by (i) the massive availability of data samples, collected in pharma driven drug programs (='big data') as well as (ii) deep learning algorithmic advancements and (iii) increase in compute power. Such applications are based on software frameworks with specific strengths and weaknesses. Here, we introduce typical applications and underlying frameworks for deep learning with a set of practical criteria for developing production ready solutions in life science and pharma research. Based on our own experience in successfully developing deep learning applications we provide suggestions and a baseline for selecting the most suited frameworks for a future-proof and cost-effective development.

Sensitivities of in vivo markers of arterial organ damage in patients with peripheral atherosclerosis

Abstract. Background: Biomarkers of vascular diseases such as ankle-brachial index (ABI), peripheral pulse pressure (pPP), central pulse pressure (cPP), and pulse wave velocity (PWV) allow assessment of arterial organ damage (AOD). However, the utility of markers other than ABI in patients with peripheral arterial disease (PAD), which are also associated with a significant increase of cardiovascular events, remains unclear. Patients and methods: Asymptomatic (n = 21) and symptomatic patients (n = 46) with a positive sonography for PAD or history of lower limb revascularization were included. ABI, pPP, cPP, and PWV were assessed. PWV were performed using a brachial cuff-based method (aortic PWV (aPWV)) and oscillography (carotid-femoral pulse wave velocity (cfPWV)), respectively. The two methods for PWV were compared using Bland Altman analysis. Sensitivities of ABI, pPP, cPP, cfPWV, and aPWV for AOD were calculated. Results: Sixty-seven patients (35.8 % female, mean age 69, range 39–91 years) had a significantly higher aPWV than cfPWV (median 10.5 m/s (IQR: 8.8–12.65 m/s) vs. median 9.0 m/s (IQR: 7.57–10.55 m/s), p = 0.0013). There was no correlation between cfPWV and age (r = 0.311, p = 0.116). Bland Altman analysis revealed a mean difference of -1.04 (-2SD; -6.38 to + 2SD; 4.31). The sensitivities for AOD were 68.7 % for ABI, 61.2 % for aPWV, 40.3 % for cfPWV, 31.3 % for peripheral PP, and 10.4 % for central aortic PP (p < 0.001). Conclusions: Brachial-derived aPWV differs from the gold standard assessment (cfPWV), which may be underestimated in PAD due to atherosclerotic obstructions along the aorto-iliac segment. The sensitivities of noninvasive in vivo markers of AOD vary widely and tend to underestimate the actual presence of AOD.

Effect of Early Prophylactic High-Dose Recombinant Human Erythropoietin in Very Preterm Infants on Neurodevelopmental Outcome at 2 Years: A Randomized Clinical Trial

IMPORTANCE

Very preterm infants are at risk of developing encephalopathy of prematurity and long-term neurodevelopmental delay. Erythropoietin treatment is neuroprotective in animal experimental and human clinical studies.

OBJECTIVE

To determine whether prophylactic early high-dose recombinant human erythropoietin (rhEPO) in preterm infants improves neurodevelopmental outcome at 2 years' corrected age.

DESIGN, SETTING, AND PARTICIPANTS

Preterm infants born between 26 weeks 0 days' and 31 weeks 6 days' gestation were enrolled in a randomized, double-blind, placebo-controlled, multicenter trial in Switzerland between 2005 and 2012. Neurodevelopmental assessments at age 2 years were completed in 2014.

INTERVENTIONS

Participants were randomly assigned to receive either rhEPO (3000 IU/kg) or placebo (isotonic saline, 0.9%) intravenously within 3 hours, at 12 to 18 hours, and at 36 to 42 hours after birth.

MAIN OUTCOMES AND MEASURES

Primary outcome was cognitive development assessed with the Mental Development Index (MDI; norm, 100 [SD, 15]; higher values indicate better function) of the Bayley Scales of Infant Development, second edition (BSID-II) at 2 years corrected age. The minimal clinically important difference between groups was 5 points (0.3 SD). Secondary outcomes were motor development (assessed with the Psychomotor Development Index), cerebral palsy, hearing or visual impairment, and anthropometric growth parameters.

RESULTS

Among 448 preterm infants randomized (mean gestational age, 29.0 [range, 26.0-30.9] weeks; 264 [59%] female; mean birth weight, 1210 [range, 490-2290] g), 228 were randomized to rhEPO and 220 to placebo. Neurodevelopmental outcome data were available for 365 (81%) at a mean age of 23.6 months. In an intention-to-treat analysis, mean MDI was not statistically significantly different between the rhEPO group (93.5 [SD, 16.0] [95% CI, 91.2 to 95.8]) and the placebo group (94.5 [SD, 17.8] [95% CI, 90.8 to 98.5]) (difference, -1.0 [95% CI, -4.5 to 2.5]; P = .56). No differences were found between groups in the secondary outcomes.

CONCLUSIONS AND RELEVANCE

Among very preterm infants who received prophylactic early high-dose rhEPO for neuroprotection, compared with infants who received placebo, there were no statistically significant differences in neurodevelopmental outcomes at 2 years. Follow-up for cognitive and physical problems that may not become evident until later in life is required.

TRIAL REGISTRATION

clinicaltrials.gov Identifier: NCT00413946.

Single-Cell Phenotype Classification Using Deep Convolutional Neural Networks

Deep learning methods are currently outperforming traditional state-of-the-art computer vision algorithms in diverse applications and recently even surpassed human performance in object recognition. Here we demonstrate the potential of deep learning methods to high-content screening-based phenotype classification. We trained a deep learning classifier in the form of convolutional neural networks with approximately 40,000 publicly available single-cell images from samples treated with compounds from four classes known to lead to different phenotypes. The input data consisted of multichannel images. The construction of appropriate feature definitions was part of the training and carried out by the convolutional network, without the need for expert knowledge or handcrafted features. We compare our results against the recent state-of-the-art pipeline in which predefined features are extracted from each cell using specialized software and then fed into various machine learning algorithms (support vector machine, Fisher linear discriminant, random forest) for classification. The performance of all classification approaches is evaluated on an untouched test image set with known phenotype classes. Compared to the best reference machine learning algorithm, the misclassification rate is reduced from 8.9% to 6.6%.

Incidence of metachronous contralateral inguinal hernias in children following unilateral repair - A meta-analysis of prospective studies

PURPOSE

The objective of this review was to systematically evaluate the incidence of a metachronous contralateral inguinal hernia (MCIH) in children with unilateral inguinal hernia and therefore to propose or to reject routine contralateral groin exploration.

METHODS

Electronic searches restricted to prospective studies with a minimal follow-up of 1year included MEDLINE, EMBASE and the Cochrane Central Register of Controlled Trials.

RESULTS

Six studies involving 1669 children were included. Overall MCIH was 6% (95% CI from 4% to 8%). The odds for MCIH development were significantly larger in children with an initial left-sided hernia (OR 2.66 with 95% CI from 1.56 to 4.53) and in children with open contralateral processus vaginalis (CPV) (OR 4.17 with 95% CI from 1.25 to 13.9).

CONCLUSIONS

The overall incidence of MCIH following unilateral inguinal hernia repair in children is 6%. Initial left-sided hernia (8.5%) and open CPV (13.8%) are risk factors for MCIH development. Female gender (8.2%) and younger age (<1year) (6.9%) non-significantly increase the risk of MCIH.

Ultrasound and Clinical Predictors of Recurrent Ischemia in Symptomatic Internal Carotid Artery Occlusion

Background and Purpose—

Occlusion of the internal carotid artery puts patients at risk of recurrent ischemic events because of hemodynamic compromise. Our goal was to characterize clinical and duplex parameters indicating patients at risk of recurrent ischemia.

Methods—

We retrospectively identified patients with symptomatic internal carotid artery occlusion. Clinical characteristics and ultrasound parameters, including collateral networks, were analyzed. Predictors for recurrent ipsilateral ischemia were investigated by Cox regression analysis.

Results—

Of 68 patients, at least 1 recurrent ischemic event within the same vascular territory was observed in 14 patients (20.6%) within 2 to 92 days (median, 29.5 days). The median follow-up period was 6 months. Diabetes mellitus and previous transient ischemic attack were associated with recurrence, as was activation of the maximum number of collateral pathways on transcranial ultrasound (28.6% versus 5.6%; P =0.03). Furthermore, flow in the posterior cerebral arteries was higher in patients with recurrence in ipsilateral and contralateral posterior cerebral artery P2 segments (76 IQR 37.5 versus 59, IQR 22.5 cm/s and 68, IQR 35.6 versus 52, IQR 21 cm/s; P <0.01 and 0.02).

Conclusions—

Flow increases in both posterior cerebral artery P2 segments suggest intensified compensatory efforts when other collaterals are insufficient. Together with the presence of diabetes mellitus and a history of transient ischemic attack, this duplex parameter indicates that patients with internal carotid artery are at particular risk of recurrent ischemia.

Unsupervised assessment of microarray data quality using a Gaussian mixture model

BACKGROUND

Quality assessment of microarray data is an important and often challenging aspect of gene expression analysis. This task frequently involves the examination of a variety of summary statistics and diagnostic plots. The interpretation of these diagnostics is often subjective, and generally requires careful expert scrutiny.

RESULTS

We show how an unsupervised classification technique based on the Expectation-Maximization (EM) algorithm and the naïve Bayes model can be used to automate microarray quality assessment. The method is flexible and can be easily adapted to accommodate alternate quality statistics and platforms. We evaluate our approach using Affymetrix 3' gene expression and exon arrays and compare the performance of this method to a similar supervised approach.

CONCLUSION

This research illustrates the efficacy of an unsupervised classification approach for the purpose of automated microarray data quality assessment. Since our approach requires only unannotated training data, it is easy to customize and to keep up-to-date as technology evolves. In contrast to other "black box" classification systems, this method also allows for intuitive explanations.

Quality assessment of Affymetrix GeneChip data

Affymetrix GeneChips are one of the best established microarray platforms. This powerful technique allows users to measure the expression of thousands of genes simultaneously. However, a microarray experiment is a sophisticated and time consuming endeavor with many potential sources of unwanted variation that could compromise the results if left uncontrolled. Increasing data volume and data complexity have triggered growing concern and awareness of the importance of assessing the quality of generated microarray data. In this review, we give an overview of current methods and software tools for quality assessment of Affymetrix GeneChip data. We focus on quality metrics, diagnostic plots, probe-level methods, pseudo-images, and classification methods to identify corrupted chips. We also describe RNA quality assessment methods which play an important role in challenging RNA sources like formalin embedded biopsies, laser-micro dissected samples, or single cells. No wet-lab methods are discussed in this paper.

RACE: Remote Analysis Computation for gene Expression data

The Remote Analysis Computation for gene Expression data (RACE) suite is a collection of bioinformatics web tools designed for the analysis of DNA microarray data. RACE performs probe-level data preprocessing, extensive quality checks, data visualization and data normalization for Affymetrix GeneChips. In addition, it offers differential expression analysis on normalized expression levels from any array platform. RACE estimates the false discovery rates of lists of potentially regulated genes and provides a Gene Ontology-term analysis tool for GeneChip data to support the biological interpretation and annotation of results. The analysis is fully automated but can be customized by flexible parameter settings. To offer a convenient starting point for subsequent analyses, and to provide maximum transparency, the R scripts used to generate the results can be downloaded along with the output files. RACE is freely available for use at http://race.unil.ch.

Multiparameter microscopy and spectroscopy for single-molecule analytics

The ability to monitor several parameters simultaneously from distinct individual fluorescent reporter molecules facilitates the disentanglement of complex and interacting systems and opens new perspectives in areas from basic science to biopharmaceutical technology. By combining annular illumination microscopy, time-correlated single-photon counting, and multichannel detection, we were able to determine 14 independent parameters from one individual fluorophore. The whole set of parameters was deduced from the few properties of the fluorescence photons, i.e., arrival time, wavelength, and polarization. With this approach, the intensity, the polarization, and the spectral dynamics can be analyzed on a nanosecond time scale and the mean values can be monitored with submillisecond time resolution. Nanosecond spectral dynamics of single molecules has been observed, to the best of our knowledge, for the first time. From our experience, we can determine all parameters for more than 30% of the illuminated fluorophores in biological samples and for more than 80% in doped polymeric films.

The citrate carrier CitS probed by single-molecule fluorescence spectroscopy

A prominent region of the Na(+)-dependent citrate carrier (CitS) from Klebsiella pneumoniae is the highly conserved loop X-XI, which contains a putative citrate binding site. To monitor potential conformational changes within this region by single-molecule fluorescence spectroscopy, the target cysteines C398 and C414 of the single-Cys mutants (CitS-sC398, CitS-sC414) were selectively labeled with the thiol-reactive fluorophores AlexaFluor 546/568 C(5) maleimide (AF(546), AF(568)). While both single-cysteine mutants were catalytically active citrate carriers, labeling with the fluorophore was only tolerated at C398. Upon citrate addition to the functional protein fluorophore conjugate CitS-sC398-AF(546), complete fluorescence quenching of the majority of molecules was observed, indicating a citrate-induced conformational change of the fluorophore-containing domain of CitS. This quenching was specific for the physiological substrate citrate and therefore most likely reflecting a conformational change in the citrate transport mechanism. Single-molecule studies with dual-labeled CitS-sC398-AF(546/568) and dual-color detection provided strong evidence for a homodimeric association of CitS.

Coupled rotation within single F0F1 enzyme complexes during ATP synthesis or hydrolysis

F0F1 ATP synthases are the smallest rotary motors in nature and work as ATP factories in bacteria, plants and animals. Here we report on the first observation of intersubunit rotation in fully coupled single F0F1 molecules during ATP synthesis or hydrolysis. We investigate the Na+-translocating ATP synthase of Propionigenium modestum specifically labeled by a single fluorophore at one c subunit using polarization-resolved confocal microscopy. Rotation during ATP synthesis was observed with the immobilized enzyme reconstituted into proteoliposomes after applying a diffusion potential, but not with a Na+ concentration gradient alone. During ATP hydrolysis, stepwise rotation of the labeled c subunit was found in the presence of 2 mM NaCl, but not without the addition of Na+ ions. Moreover, upon the incubation with the F0-specific inhibitor dicyclohexylcarbodiimide the rotation was severely inhibited.

Probing confined fields with single molecules and vice versa

Single dye molecules are used as local probes to map the spatial distribution of the squared electric field components in the focus of a high numerical aperture lens. Simulated field distributions are quantitatively verified by experimentally obtained fluorescence excitation maps. We show that annular illumination can be used to engineer the field distribution in the focus at a dielectric/air interface such that electric field components in all directions acquire comparable magnitudes. The 3D orientation of molecular absorption dipoles can be determined by comparing measured to simulated image patterns. The presence of longitudinal electric field components in a focus is of particular interest in tip-enhanced scanning near-field optical microscopy.

Orientational imaging of single molecules by annular illumination

The absorption dipole orientation of single fluorescent molecules is determined by mapping the spatial distribution of the squared electric field components in a high-numerical-aperture laser focus. Annular illumination geometry and the vicinity of a plane dielectric/air interface strongly enhance the longitudinal field component and the transverse fields perpendicular to the polarization direction. As a result, all three excitation field components in the focus are of comparable magnitude. The scheme holds promise to monitor rotational diffusion of single molecules in complex environments.

Wave-front reconstruction with a shack-hartmann sensor with an iterative spline fitting method

One limitation of the conventional Shack-Hartmann sensor is that the spots of each microlens have to remain in their respective subapertures. We present an algorithm that assigns the spots to their reference points unequivocally even if they are situated far outside their subaperture. For this assignment a spline function is extrapolated in successive steps of the iterative algorithm. The proposed method works in a single-shot technique and does not need any aid from mechanical devices. The reconstruction of a simulated steep aspherical wave front (approximately 100 lambda/mm slope) is described as well as experimental results of the measurement of a spherical wave front with a huge peak-to-valley value (approximately 400 lambda). The performance of the method is compared with the unwrapping method, which has been published before.

Optical Microscopy in the Nano-World

Scanning near-field optical microscopy (SNOM) is an optical microscopy whose resolution is not bound to the diffraction limit. It provides chemical information based upon spectral, polarization and/or fluorescence contrast images. Details as small as 20 nm can be recognized. Photophysical and photochemical effects can be studied with SNOM on a similar scale. This article reviews a good deal of the experimental and theoretical work on SNOM in Switzerland.