Background/Objectives: This study aimed to create a pipeline for standardized data extraction from lumbar-spine MRI radiology reports using a large language model (LLM) and assess the agreement of the extracted data with research-grade semi-quantitative scoring. Methods: We included a subset of data from a multi-site NIH-funded cohort study of chronic low back pain (cLBP) participants. After initial prompt development, a secure application programming interface (API) deployment of OpenAIs GPT-4 was used to extract different classes of pathology from the clinical radiology report. Unsupervised UMAP and agglomerative clustering of the pathology terms’ embeddings provided insight into model comprehension for optimized prompt design. Model extraction was benchmarked against human extraction (gold standard) with F1 scores and false-positive and false-negative rates (FPR/FNR). Then, an expert MSK radiologist provided comprehensive research-grade scores of the images, and agreement with report-extracted data was calculated using Cohen’s kappa. Results: Data from 230 patients with cLBP were included (mean age 53.2 years, 54% women). The overall model performance for extracting data from clinical reports was excellent, with a mean F1 score of 0.96 across pathologies. The mean FPR was marginally higher than the FNR (5.1% vs. 3.0%). Agreement with comprehensive scoring was moderate (kappa 0.424), and the underreporting of lateral recess stenosis (FNR 63.6%) and overreporting of disc pathology (FPR 42.7%) were noted. Conclusions: LLMs can accurately extract highly detailed information on lumbar spine imaging pathologies from radiology reports. Moderate agreement between the LLM and comprehensive scores underscores the need for less subjective, machine-based data extraction from imaging.

Entheses have been discussed as potential sources of pain in osteoarthritis (OA). Entheseal bone remodeling quantified by [18F]NaF-PET could be a surrogate marker for tensile forces causing a painful tissue response. This study aimed to investigate the relationship between [18 F]NaF-uptake, MRI findings and knee-related symptoms. Patients and healthy controls in this prospective observational study on patello-femoral OA (PFJOA) underwent simultaneous MR (at 3.0 T) and [18F]NaF PET imaging of both knees. Cartilage damage, subchondral marrow lesions, synovitis, and enthesophytes were semi-quantitatively assessed on MRI and entheseal tracer uptake was quantified in the anterior patella and tibial tuberosity. Symptoms were assessed with the Knee injury and Osteoarthritis Outcome Score (KOOS). Generalized estimating equations were used to investigate associations between tracer uptake and symptoms; analyses were additionally adjusted for MRI-detected changes (as alternative symptom drivers), age, sex and body mass index. A total of 135 knees from 68 subjects were investigated (mean age 48.8 ± 10.4 years; 58.8% women). Higher [18F]NaF-uptake was significantly associated with worse outcomes for KOOS subscales sports (beta -2.7; p<0.001), quality of live (beta 3.0; p<0.001) and PFJ (beta 2.3; p=0.001), but not pain (beta -0.8; p=0.18), symptoms (beta -0.6; p=0.30) or activities of daily living (beta -1.1; p=0.06). Entheseal bone remodeling quantified by [18F]NaF-PET-MRI is associated with patient-reported clinical knee outcomes. These findings underscore the potential of entheseal biology to inform the understanding of symptom generation in degenerative disease of the musculoskeletal system.

To study the longitudinal changes of cartilage ${\mathrm{T}}_{1\mathrm{\rho }}$ and ${\mathrm{T}}_2$ relaxation time measurements in hip-OA patients. A calibration study compared two scanner data, Scanner-1 (GE Discovery MR750 3.0T) with unilateral acquisition protocol and Scanner-2 (GE Signa Premier 3.0T) with bilateral acquisition protocol, using nine subjects(average age ​= ​40.33 ​± ​13.53 years, 5 females), including one hip-OA subject. Quantified parameters from the Scanner-2 were adjusted using voxel-based relaxometry(VBR) and Z-score normalization to reduce the inter-scanner variability. Eighteen hip-OA Subjects (age ​= ​53.11 ​± ​14.96 years, 12 females) were recruited to the longitudinal variability study from 2016, comprising five assessments at 1-year intervals. Baseline to 3rd-year data used unilateral acquisition with Scanner-1, while 4th-year data used bilateral acquisition with Scanner-2. A linear mixed-effects model(LME) assessed trajectory analyses, with acquisition year, age, sex, body mass index(BMI), and Kellgren-Lawrence(KL) score as predictor variables and cartilage mean ${\mathrm{T}}_{1\mathrm{\rho }}$ and ${\mathrm{T}}_2$ values as outcomes. VBR analysis after Z-score normalization showed that only a few of the whole cartilage voxels had significant differences in ${\mathrm{T}}_{1\mathrm{\rho }}($femur-2.36 ​% and acetabular-3.23 ​%) and ${\mathrm{T}}_2$ (femur-2.30 ​% and acetabular-2.94 ​%) values between the scanners. The LME analysis showed that the BMI predictor variable was significantly correlated with the femur ${\mathrm{T}}_{1\mathrm{\rho }}$ (p ​< ​0.0001) and ${\mathrm{T}}_2$ (p ​< ​0.0001) and acetabular ${\mathrm{T}}_{1\mathrm{\rho }}$ (p ​< ​0.0001) and ${\mathrm{T}}_2$ (p ​< ​0.0001) cartilage region. The calibration study demonstrated the effectiveness of VBR and Z-score normalization in reducing inter-scanner variability. The longitudinal study revealed a significant correlation between ${\mathrm{T}}_{1\mathrm{\rho }}$ and ${\mathrm{T}}_2$ values of the cartilage and BMI; also the ${\mathrm{T}}_{1\mathrm{\rho }}$ and ${\mathrm{T}}_2$ values increased over time in some of the cartilage subregions.