Performance of an Open-Source Large Language Model in Extracting Information from Free-Text Radiology Reports.

Purpose To assess the performance of a local open-source large language model (LLM) in various information extraction tasks from real-life emergency brain MRI reports. Materials and Methods All consecutive emergency brain MRI reports written in 2022 from a French quaternary center were retrospectively reviewed. Two radiologists identified MRI scans that were performed in the emergency department for headaches. Four radiologists scored the reports' conclusions as either normal or abnormal. Abnormalities were labeled as either headache-causing or incidental. Vicuna (LMSYS Org), an open-source LLM, performed the same tasks. Vicuna's performance metrics were evaluated using the radiologists' consensus as the reference standard. Results Among the 2398 reports during the study period, radiologists identified 595 that included headaches in the indication (median age of patients, 35 years [IQR, 26-51 years]; 68% [403 of 595] women). A positive finding was reported in 227 of 595 (38%) cases, 136 of which could explain the headache. The LLM had a sensitivity of 98.0% (95% CI: 96.5, 99.0) and specificity of 99.3% (95% CI: 98.8, 99.7) for detecting the presence of headache in the clinical context, a sensitivity of 99.4% (95% CI: 98.3, 99.9) and specificity of 98.6% (95% CI: 92.2, 100.0) for the use of contrast medium injection, a sensitivity of 96.0% (95% CI: 92.5, 98.2) and specificity of 98.9% (95% CI: 97.2, 99.7) for study categorization as either normal or abnormal, and a sensitivity of 88.2% (95% CI: 81.6, 93.1) and specificity of 73% (95% CI: 62, 81) for causal inference between MRI findings and headache. Conclusion An open-source LLM was able to extract information from free-text radiology reports with excellent accuracy without requiring further training. Keywords: Large Language Model (LLM), Generative Pretrained Transformers (GPT), Open Source, Information Extraction, Report, Brain, MRI Supplemental material is available for this article. Published under a CC BY 4.0 license. See also the commentary by Akinci D'Antonoli and Bluethgen in this issue.

Effects of a 20-Week High-Intensity Strength Training Program on Muscle Strength Gain and Cardiac Adaptation in Untrained Men: Preliminary Results of a Prospective Longitudinal Study.

BACKGROUND: As strength sports gain popularity, there is a growing need to explore the impact of sustained strength training on cardiac biventricular structure and function, an area that has received less attention compared to the well-established physiological cardiac adaptation to endurance training. OBJECTIVE: This study aims to implement a 20-week high-intensity strength training program to enhance maximal muscle strength and evaluate its impact on cardiac biventricular adaptation in healthy, untrained men. METHODS: A total of 27 healthy and untrained young men (mean age 22.8, SD 3.2 years) participated in a strength training program designed to increase muscle strength. The training program involved concentric, eccentric, and isometric exercise phases, conducted over a consecutive 20-week time frame with a frequency of 3 weekly training sessions. Participants were evaluated before and after 12 and 20 weeks of training through body composition analysis (bioelectrical impedance), a 12-lead resting electrocardiogram, 3D transthoracic echocardiography, cardiopulmonary exercise testing, and muscle isokinetic dynamometry. The progression of strength training loads was guided by 1-repetition maximum (RM) testing during the training program. RESULTS: Of the initial cohort, 22 participants completed the study protocol. No injuries were reported. The BMI (mean 69.8, SD 10.8 kg/m² vs mean 72, SD 11 kg/m²; P=.72) and the fat mass (mean 15.3%, SD 7.5% vs mean 16.5%, SD 7%; P=.87) remained unchanged after training. The strength training program led to significant gains in 1-RM exercise testing as early as 4 weeks into training for leg extension (mean 69.6, SD 17.7 kg vs mean 96.5, SD 31 kg; P<.001), leg curl (mean 43.2, SD 9.7 kg vs mean 52.8, SD 13.4 kg; P<.001), inclined press (mean 174.1, SD 41.1 kg vs mean 229.2, SD 50.4 kg; P<.001), butterfly (mean 26.3, SD 6.2 kg vs mean 32.5, SD 6.6 kg; P<.001), and curl biceps on desk (mean 22.9, SD 5.2 kg vs mean 29.6, SD 5.2 kg; P<.001). After 20 weeks, the 1-RM leg curl, bench press, pullover, butterfly, leg extension, curl biceps on desk, and inclined press showed significant mean percentage gains of +40%, +41.1%, +50.3%, +63.5%, +80.1%, +105%, and +106%, respectively (P<.001). Additionally, the isokinetic evaluation confirmed increases in maximal strength for the biceps (+9.2 Nm), triceps (+11.6 Nm), quadriceps (+46.8 Nm), and hamstrings (+25.3 Nm). In this paper, only the training and muscular aspects are presented; the cardiac analysis will be addressed separately. CONCLUSIONS: This study demonstrated that a short-term high-intensity strength training program was successful in achieving significant gains in muscle strength among previously untrained young men. We intend to use this protocol to gain a better understanding of the impact of high-intensity strength training on cardiac physiological remodeling, thereby providing new insights into the cardiac global response in strength athletes. TRIAL REGISTRATION: ClinicalTrials.gov NCT04187170; https://clinicaltrials.gov/study/NCT04187170.

Anomalous nonlinear response of glassy liquids: general arguments and a mode-coupling approach.

We study theoretically the nonlinear response properties of glass formers. We establish several general results, which together with the assumption of time-temperature superposition, lead to a relation between the nonlinear response and the derivative of the linear response with respect to temperature. Using results from mode-coupling theory and scaling arguments valid close to the glass transition, we obtain the frequency and temperature dependence of the nonlinear response in the alpha- and beta-regimes. Our results demonstrate that supercooled liquids are characterized by responses to external perturbations that become increasingly nonlinear as the glass transition is approached. These results are extended to the case of inhomogeneous perturbing fields.

Field theories and exact stochastic equations for interacting particle systems.

We consider the dynamics of interacting particles with reaction and diffusion. Starting from the underlying discrete stochastic jump process we derive a general field theory describing the dynamics of the density field, which we relate to an exact stochastic equation on the density field. We show how our field theory maps onto the original Doi-Peliti formalism, allowing us to clarify further the issue of the "imaginary" Langevin noise that appears in the context of reaction-diffusion processes. Our procedure applies to a wide class of problems and is related to large deviation functional techniques developed recently to describe fluctuations of nonequilibrium systems in the hydrodynamic limit.

Spatially heterogeneous dynamics in a model for granular compaction.

We suggest the emergence of spatially correlated dynamics in slowly compacting dense granular media by analyzing analytically and numerically multipoint correlation functions in a simple particle model characterized by slow nonequilibrium dynamics. We show that the logarithmically slow dynamics at large times is accompanied by spatially extended dynamic structures that resemble the ones observed in glass-forming liquids and dense colloidal suspensions. This suggests that dynamic heterogeneity is another key common feature present in very different jamming materials.

Possible test of the thermodynamic approach to granular media.

We study the steady state distribution of the energy of the Sherrington-Kirkpatrick model driven by a tapping mechanism which mimics the mechanically driven dynamics of granular media. The dynamics consists of two phases: a zero temperature relaxation phase which leads the system to a metastable state, then a tapping which excites the system thus reactivating the relaxational dynamics. Numerically, we investigate whether the distribution of the energies of the blocked states obtained agrees with a simple canonical form of the Edwards measure. It is found that this canonical measure is in good agreement with the dynamically measured energy distribution. A possible experimental test of the Edwards measure based on the study here is proposed.