Application of a Domain-specific BERT for Detection of Speech Recognition Errors in Radiology Reports.

Purpose: To develop radiology domain-specific bidirectional encoder representations from transformers (BERT) models that can identify speech recognition (SR) errors and suggest corrections in radiology reports. Materials and Methods: A pretrained BERT model, Clinical BioBERT, was further pretrained on a corpus of 114 008 radiology reports between April 2016 and August 2019 that were retrospectively collected from two hospitals. Next, the model was fine-tuned on a training dataset of generated insertion, deletion, and substitution errors, creating Radiology BERT. This model was retrospectively evaluated on an independent dataset of radiology reports with generated errors (n = 18 885) and on unaltered report sentences (n = 2000) and prospectively evaluated on true clinical SR errors (n = 92). Correction Radiology BERT was separately trained to suggest corrections for detected deletion and substitution errors. Area under the receiver operating characteristic curve (AUC) and bootstrapped 95% CIs were calculated for each evaluation dataset. Results: Radiology-specific BERT had AUC values of >.99 (95% CI: >0.99, >0.99), 0.94 (95% CI: 0.93, 0.94), 0.98 (95% CI: 0.98, 0.98), and 0.97 (95% CI: 0.97, 0.97) for detecting insertion, deletion, substitution, and all errors, respectively, on the independently generated test set. Testing on unaltered report impressions revealed a sensitivity of 82% (28 of 34; 95% CI: 70%, 93%) and specificity of 88% (1521 of 1728; 95% CI: 87%, 90%). Testing on prospective SR errors showed an accuracy of 75% (69 of 92; 95% CI: 65%, 83%). Finally, the correct word was the top suggestion for 45.6% (475 of 1041; 95% CI: 42.5%, 49.3%) of errors. Conclusion: Radiology-specific BERT models fine-tuned on generated errors were able to identify SR errors in radiology reports and suggest corrections.Keywords: Computer Applications, Technology Assessment Supplemental material is available for this article. © RSNA, 2022See also the commentary by Abajian and Cheung in this issue.

Inferring spatial variations of microstructural properties from macroscopic mechanical response.

Disease alters tissue microstructure, which in turn affects the macroscopic mechanical properties of tissue. In elasticity imaging, the macroscopic response is measured and is used to infer the spatial distribution of the elastic constitutive parameters. When an empirical constitutive model is used, these parameters cannot be linked to the microstructure. However, when the constitutive model is derived from a microstructural representation of the material, it allows for the possibility of inferring the local averages of the spatial distribution of the microstructural parameters. This idea forms the basis of this study. In particular, we first derive a constitutive model by homogenizing the mechanical response of a network of elastic, tortuous fibers. Thereafter, we use this model in an inverse problem to determine the spatial distribution of the microstructural parameters. We solve the inverse problem as a constrained minimization problem and develop efficient methods for solving it. We apply these methods to displacement fields obtained by deforming gelatin-agar co-gels and determine the spatial distribution of agar concentration and fiber tortuosity, thereby demonstrating that it is possible to image local averages of microstructural parameters from macroscopic measurements of deformation.

Noninvasive In-Vivo Quantification of Mechanical Heterogeneity of Invasive Breast Carcinomas.

Heterogeneity is a hallmark of cancer whether one considers the genotype of cancerous cells, the composition of their microenvironment, the distribution of blood and lymphatic microvasculature, or the spatial distribution of the desmoplastic reaction. It is logical to expect that this heterogeneity in tumor microenvironment will lead to spatial heterogeneity in its mechanical properties. In this study we seek to quantify the mechanical heterogeneity within malignant and benign tumors using ultrasound based elasticity imaging. By creating in-vivo elastic modulus images for ten human subjects with breast tumors, we show that Young's modulus distribution in cancerous breast tumors is more heterogeneous when compared with tumors that are not malignant, and that this signature may be used to distinguish malignant breast tumors. Our results complement the view of cancer as a heterogeneous disease on multiple length scales by demonstrating that mechanical properties within cancerous tumors are also spatially heterogeneous.

Using an organic molecule with low triplet energy as a host in a highly efficient blue electrophosphorescent device.

To achieve high efficiencies in blue phosphorescent organic light-emitting diodes (PhOLEDs), the triplet energies (T1) of host materials are generally supposed to be higher than the blue phosphors. A small organic molecule with low singlet energy (S1) of 2.80 eV and triplet energy of 2.71 eV can be used as the host material for the blue phosphor, [bis(4,6-difluorophenylpyridinato-N,C(2'))iridium(III)] tetrakis(1-pyrazolyl)borate (FIr6; T1=2.73 eV). In both the photo- and electro-excited processes, the energy transfer from the host material to FIr6 was found to be efficient. In a three organic-layer device, the maximum current efficiency of 37 cd A(-1) and power efficiency of 40 Lm W(-1) were achieved for the FIr6-based blue PhOLEDs.

Simple bipolar molecules constructed from biphenyl moieties as host materials for deep-blue phosphorescent organic light-emitting diodes.

Simple is good! Based on biphenyl molecules, two bipolar host materials with high triplet energies have been rationally designed, synthesized, and fully characterized. Deep blue phosphorescent organic light-emitting diodes, which employ the new hosts and an iridium(III) complex as triplet emitter, show a maximum current efficiency of 40 cd A(-1), a maximum power efficiency of 36 lm W(-1), and a maximum external quantum efficiency of 19.5 %.