Highly accurate classification of chest radiographic reports using a deep learning natural language model pre-trained on 3.8 million text reports.

MOTIVATION: The development of deep, bidirectional transformers such as Bidirectional Encoder Representations from Transformers (BERT) led to an outperformance of several Natural Language Processing (NLP) benchmarks. Especially in radiology, large amounts of free-text data are generated in daily clinical workflow. These report texts could be of particular use for the generation of labels in machine learning, especially for image classification. However, as report texts are mostly unstructured, advanced NLP methods are needed to enable accurate text classification. While neural networks can be used for this purpose, they must first be trained on large amounts of manually labelled data to achieve good results. In contrast, BERT models can be pre-trained on unlabelled data and then only require fine tuning on a small amount of manually labelled data to achieve even better results. RESULTS: Using BERT to identify the most important findings in intensive care chest radiograph reports, we achieve areas under the receiver operation characteristics curve of 0.98 for congestion, 0.97 for effusion, 0.97 for consolidation and 0.99 for pneumothorax, surpassing the accuracy of previous approaches with comparatively little annotation effort. Our approach could therefore help to improve information extraction from free-text medical reports. Availability and implementationWe make the source code for fine-tuning the BERT-models freely available at https://github.com/fast-raidiology/bert-for-radiology. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Experimental and Numerical Investigation of an Axial Rotary Blood Pump.

Left ventricular assist devices (LVADs) have become a standard therapy for patients with severe heart failure. As low blood trauma in LVADs is important for a good clinical outcome, the assessment of the fluid loads inside the pump is critical. More specifically, the flow features on the surfaces where the interaction between blood and artificial material happens is of great importance. Therefore, experimental data for the near-wall flows in an axial rotary blood pump were collected and directly compared to computational fluid dynamic results. For this, the flow fields based on unsteady Reynolds-averaged Navier-Stokes simulations-computational fluid dynamics (URANS-CFD) of an axial rotary blood pump were calculated and compared with experimental flow data at one typical state of operation in an enlarged model of the pump. The focus was set on the assessment of wall shear stresses (WSS) at the housing wall and rotor gap region by means of the wall-particle image velocimetry technique, and the visualization of near-wall flow structures on the inner pump surfaces by a paint erosion method. Additionally, maximum WSS and tip leakage volume flows were measured for 13 different states of operation. Good agreement between CFD and experimental data was found, which includes the location, magnitude, and direction of the maximum and minimum WSS and the presence of recirculation zones on the pump stators. The maximum WSS increased linearly with pressure head. They occurred at the upstream third of the impeller blades and exceeded the critical values with respect to hemolysis. Regions of very high shear stresses and recirculation zones could be identified and were in good agreement with simulations. URANS-CFD, which is often used for pump performance and blood damage prediction, seems to be, therefore, a valid tool for the assessment of flow fields in axial rotary blood pumps. The magnitude of maximum WSS could be confirmed and were in the order of several hundred Pascal.