Search

Data science systems (DSSs) are a fundamental tool in many areas of research and are now developed by people with a myriad of backgrounds. This is coupled with a crisis in reproducibility of such DSSs despite the wide availability of powerful tools for data science and machine learning over the last decade. We believe that perverse incentives and a lack of widespread software engineering skills are among the many causes of this crisis and analyze why software engineering and building large complex systems is, in general, hard. Based on these insights, we identify how software engineering addresses those difficulties and how one might apply and generalize software engineering methods to make DSSs more fit for purpose. We advocate two key development philosophies: one should incrementally grow – not plan then build – DSSs, and one should use two types of feedback loops during development: one which tests the code’s correctness and another that evaluates the code’s efficacy.

Ziel: Die DRG-ÖRG-IRP (Deutsche Röntgengesellschaft-Österreichische Röntgengesellschaft Internationale Radiomics-Plattform) stellt eine web-/cloudbasierte Radiomics-Plattform auf Grundlage einer öffentlich-privaten Partnerschaft dar. Sie bietet die Möglichkeit der gemeinsamen Nutzung von Daten, Annotation, Validierung und Zertifizierung auf dem Gebiet der künstlichen Intelligenz, Radiomics-Analyse und integrierten Diagnostik. In einer ersten Proof-of-Concept-Studie soll die automatisierte Myokardsegmentation sowie die automatisierte myokardiale Late-Gadolinum-Enhancement (LGE)-Detektion mittels radiomischer Bildmerkmale für Myokarditis-Datensätze evaluiert werden. Material und Methoden: Die DRG-ÖRP-IRP kann zur Erstellung qualitätsgesicherter, strukturierter Bilddaten in Kombination mit klinischen Daten und anschließender integrierter Datenanalyse genutzt werden und zeichnet sich durch die folgenden Leistungskriterien aus: Nutzungsmöglichkeit multizentrischer vernetzter Daten, automatisiert berechnete Qualitätsparameter, Bearbeitung von Annotationsaufgaben, Konturerkennung mittels herkömmlicher Verfahren sowie Verfahren der künstlichen Intelligenz und der Möglichkeit einer gezielten Einbindung von Algorithmen. In einer ersten Studie wurde ein anhand kardialer CINE-Datensätze vortrainiertes neuronales Netz zur Segmentierung von PSIR-Datensätzen evaluiert. In einem zweiten Schritt wurden radiomische Bildmerkmale zur segmentalen Detektion von LGE der gleichen Datensätze, welche multizentrisch über die IRP zu Verfügung gestellt wurden, angewendet. Ergebnisse: Erste Ergebnisse zeigen die Vorteile (Datentransparenz, Zuverlässigkeit, breite Einbindung aller Mitglieder, kontinuierliche Evolution sowie Validierung und Zertifizierung) dieses plattformbasierten Ansatzes auf. In der Proof-of-Concept-Studie konnte im Vergleich zur Segmentierung des Myokards durch den Experten das neuronale Netzwerk einen Dice-Koeffizienten von 0,813 nachweisen. Im Rahmen der segmentbasierten myokardialen LGE-Detektion ergab sich ein AUC von 0,73 bzw. ein AUC von 0,79 nach Ausschluss von Segmenten mit unsicherer Annotation. Die Auswertung und Bereitstellung der Daten findet auf der IRP unter Berücksichtigung der FAT-Kriterien (Fairness, Accountability, Transparency) sowie FAIR-Kriterien (findable, accessible, interoperable, reusable) statt. Schlussfolgerung: Es konnte gezeigt werden, dass die DRG-ÖRP-IRP als Kristallisationspunkt für die Generierung weiterer Einzel- und Verbundprojekte genutzt werden kann. Die Durchführung von quantitativen Analysen mit Verfahren der künstlichen Intelligenz wird hierbei durch den Plattformansatz der DRG-ÖRP-IRP erheblich erleichtert, da vorab trainierte neuronale Netze integriert und wissenschaftliche Gruppen vernetzt werden können. In einer ersten Proof-of-Concept-Studie zur automatisierten Segmentation des Myokards sowie zur automatisierten myokardialen LGE-Detektion konnten diese Vorteile erfolgreich angewendet werden. Somit zeigt sich, dass sich mittels der DRG-ÖRP-IRP strategische Ziele interdisziplinär umsetzen, konkrete alltagstaugliche Proof-of-Concept-Beispiele aufzeigen sowie möglichst partizipativ unter Einbindung aller Gruppierungen eine Vielzahl an Einzel- und Verbundprojekten realisieren lassen.

Contains fulltext : 100958.pdf (Author’s version preprint ) (Open Access) Covert attention has been introduced as a paradigm for gaze-independent brain computer interfacing (BCI). As we know that the applicability of a BCI system depends on the consistency of its paradigm over a group of subjects, one important feature of a BCI is its subject-independence. To generalize a BCI to be applicable for new users, one can think of either to use a machine learning algorithm above a standard BCI to improve the generalization or to choose a more appropriate paradigm. Various machine learning algorithms have been applied on imaginary movement based paradigms towards having a subject-independent BCI, yet there are a few efforts on testing other paradigms in this context. In this paper we show that covert attention can be used as a robust paradigm for subject-independent brain computer interfacing.

Fragestellung: Der derzeitige Gold-standard zur operativen Behandlung der instabilen distalen Radiusfraktur ist die volare Plattenosteosynthese mittels anatomisch prä-konfektionierten Platten. Dabei variiert der Schaft-volare Fläche Winkel (VCA) dieser Platten zwischen 15 und 25°. Aktuell[zum vollständigen Text gelangen Sie über die oben angegebene URL], Deutscher Kongress für Orthopädie und Unfallchirurgie (DKOU 2017)

Contains fulltext : 100980.pdf (author's version ) (Closed access) As the number of possible neural sources is much higher than the number of MEG or EEG sensor readings, the inverse problem of estimating source amplitudes from sensor readings has many solutions. A common approach to tackle this problem is to assume that all sources are independent from each other. This approach is widely used in the neuroscience community and is known as beamforming. Since the source amplitude is likely to change smoothly over time, we expect to improve the source localization by taking the temporal dynamics into account. In this paper, we incorporate the independence assumption of the standard beamformer in a linear dynamic system, and we show that by using the leadfield matrix as the observation model and setting the covariance of the observation noise to be proportional to the covariance of the observation, we arrive at the dynamic beamformer International Workshop on Machine Learning and Interpretation in Neuroimaging (MLINI 2011) (Sierra Nevada, Spain, December 16-17, 2011)

In radiation therapy, multiatlas segmentation is recognized as being accurate, but is generally not considered scalable since the highest accuracy is achieved only when using a large atlas database. The fundamental problem is to use such a large database, to accurately represent the population variability, while conserving a relatively small computational cost. A method based on the composition of transformations is proposed to address this issue. The main novelties and key contributions of this paper are the definition of a transitivity error function and the presentation of an image clustering scheme that is based solely on the computed registration transformations. Leave-one-out experiments conducted on a database of N = 50 MR prostate scans demonstrate that a reduction of (N - 1) = 49x in the number of pre-alignment registrations, and of 3.2x in term of total registration effort, is possible without significant impact on segmentation quality.

Objective. Comparison of performances of 5 (semi)automated methods in measuring joint space width (JSW) in rheumatoid arthritis. Methods. Change in JSW was determined by 5 measurement methods on 4 radiographs per patient from 107 patients included in the COBRA trial (comparing sulfasalazine alone or in combination with methotrexate and corticosteroids). For each method the number of patients with sufficient available results was assessed (efficiency). An independent repeated measurement was carried out on a random sample of 30 patients' baseline and 1-year radiographs, to evaluate within-method reliability of change scores. Discriminatory ability (DA) of the measurement methods (between the 2 treatment arms) was compared with the DA of the Sharp-van der Heijde score (SHS) and its 2 components (erosion and JSW scores). Results. The overall success rate varied widely between methods. Applying the chosen threshold of a minimum of 50% available joints with a change score per patient resulted in a success rate > 92% in 4/5 methods. Repeatability of measurements was good for most methods (intraclass correlation coefficient ≥ 0.80 in 4/5 methods). Almost all measurement methods in 3 followup periods (12/14) showed a lower mean loss of JSW in patients from the intensive treatment group, although this was rarely statistically significant, confirming the known difference in structural damage. JSW as measured by the (semi)automated systems often showed higher DA than the JSW score of the SHS, but was lower than the total SHS and erosion scores. Conclusion. Although efficiency of the methods should be improved further, results already show good reliability and encouraging DA of most methods. Optimal information may be obtained with a combination of scoring of erosions and (semi)automated measurement of JSW. (First Release June 15 2008; J Rheumatol 2008;35:1288-93)

Computer-based methods of measuring joint space width (JSW) could potentially have advantages over scoring joint space narrowing, with regard to increased standardization, sensitivity, and reproducibility. In an early exercise, 4 different methods showed good agreement on measured change in JSW over time in the small joints of the hands and feet. Despite differences in measurement values between methods, measurement of within-joint change over time showed no systematic differences. The within-method variation was small, with intra-operator variation being smaller than inter-operator variation. Although this initial study was limited in terms of the number of patients and timepoints (total 10), the number of joints was relatively high (340 joints), so the results were considered strong evidence supporting the validity of computer-based JSW measurements to continue the study of the potential value of JSW by comparison of measurements to manual scoring of joint space narrowing using the COBRA trial images.