Exploring the generative capabilities of the multimodal GPT-4, our study uncovered significant differences between radiological assessments and automatic evaluation metrics for chest x-ray impression generation and revealed radiological bias.

Background: Shoulder dislocations represent common injuries and are often combined with rotator cuff tears and potentially damage to the biceps pulley. Purpose: To assess the occurrence and type of biceps pulley lesions in patients after traumatic anterior shoulder dislocation using 3T MRI. Methods: Thirty-three consecutive patients were enrolled between June 2021 and March 2022 (14 women, mean age 48.0 ± 19 years). All patients underwent MR imaging at 3 T within one week. Images were analyzed for the presence and type of pulley tears, subluxation/dislocation of the LHBT, rotator cuff lesions, joint effusion, labral lesions, and osseous defects. Results: Seventeen patients (52%) with traumatic anterior shoulder dislocation demonstrated biceps pulley lesions. Of those, eleven patients (33%) showed a combined tear of the sGHL and CHL. All seventeen patients with lesions of the biceps pulley showed associated partial tearing of the rotator cuff, whereas three patients showed an additional subluxation of the LHBT. Patients with pulley lesions after dislocations were significantly older than those without (mean age 52 ± 12 years vs. 44 ± 14 years, p = 0.023). Conclusion: Our results suggest an increased awareness for lesions of the biceps pulley in acute traumatic shoulder dislocation, particularly in patients over 45 years.

Vertebral fracture grading classifies the severity of vertebral fractures, which is a challenging task in medical imaging and has recently attracted Deep Learning (DL) models. Only a few works attempted to make such models human-interpretable despite the need for transparency and trustworthiness in critical use cases like DL-assisted medical diagnosis. Moreover, such models either rely on post-hoc methods or additional annotations. In this work, we propose a novel interpretable-by-design method, ProtoVerse, to find relevant sub-parts of vertebral fractures (prototypes) that reliably explain the model’s decision in a human-understandable way. Specifically, we introduce a novel diversity-promoting loss to mitigate prototype repetitions in small datasets with intricate semantics. We have experimented with the VerSe’19 dataset and outperformed the existing prototype-based method. Further, our model provides superior interpretability against the post-hoc method. Importantly, expert radiologists validated the visual interpretability of our results, showing clinical applicability.

UNSTRUCTURED Exploring the generative capabilities of the multimodal GPT-4, our study uncovered significant differences between radiological assessments and automatic evaluation metrics for chest x-ray impression generation and revealed radiological bias.

Background/Objectives: Lung nodules present a common diagnostic challenge, particularly when benign and malignant lesions exhibit similar imaging characteristics. Standard evaluation relies on computed tomography (CT), positron emission tomography (PET), or biopsy, all of which have limitations. Quantitative magnetic resonance (MR) relaxometry using native longitudinal relaxation time (T1) and transverse relaxation time (T2) mapping offers a radiation-free alternative reflecting tissue-specific differences. Methods: This prospective, single-center study included 64 patients with 76 histologically or radiologically confirmed lung lesions (25 primary lung cancers, 28 metastases, 9 granulomas, and 14 pneumonic infiltrates). The patients underwent T1 and T2 mapping at 3T. Two independent readers quantified the mean values for each lesion. The pre-specified primary endpoints were (1) benign versus malignant and (2) primary lung cancer versus pulmonary metastases. Results: Significant differences in T1 and T2 values were observed across lesion types. Benign lesions exhibited high T2 values (mean 213.6 ms) and low T1 values (mean 836.6 ms), whereas malignant tumors exhibited lower T2 values (~77–78 ms) and higher T1 values (~1460–1504 ms, p < 0.001). Binary classification yielded 95.7% accuracy (sensitivity 93.8% for malignant, specificity 100% for benign) in an internal 70/30 hold-out validation (no external dataset), with consistent performance confirmed by patient-level and nested cross-validation (balanced accuracy ≈ 0.92–0.94). However, malignant subtypes could not be reliably distinguished (p > 0.05), and multiclass accuracy was 60.9%. Conclusions: Quantitative MR relaxometry allows accurate, radiation-free differentiation of benign and malignant lung lesions and may help reduce unnecessary invasive procedures.

Background/Objectives: To assess associations between quantitative computed tomography (CT) features, inflammatory markers, and length of hospital stay (LOS) in acute pyelonephritis (APN). Methods: This retrospective single-center study included 82 patients with CT-confirmed APN. Two radiologists quantified renal perfusion-deficit volume and percentage and perirenal fat-stranding (PFS) area and attenuation. Associations with C-reactive protein (CRP), white blood cell (WBC) count, procalcitonin, and LOS were assessed using Spearman correlation and regression analyses. Receiver operating characteristic (ROC) curve analysis evaluated LOS ≥ 10 days. Results: Perfusion-deficit volume and percentage correlated strongly with CRP (ρ = 0.763 and 0.714; both p < 0.001) and weakly with WBC count (ρ = 0.379 and 0.374; both p < 0.001). PFS area correlated moderately with procalcitonin (ρ = 0.482, p = 0.001; n = 42). Both perfusion-deficit measures correlated moderately with LOS (ρ = 0.538 and 0.536; both p < 0.001). In adjusted linear regression, CRP remained associated with LOS (Coefficient = 0.523 days per 10 mg/L; 95% CI, 0.345–0.692; p < 0.001), whereas perfusion-deficit percentage did not (p = 0.430). In adjusted logistic regression, CRP, age, and male sex were associated with LOS ≥ 10 days. The multivariable model yielded an AUC of 0.883 (95% CI, 0.81–0.95). Results were unchanged in a sensitivity analysis excluding the seven outpatients, and bootstrap internal validation yielded an optimism-corrected AUC of 0.86. Conclusions: Quantitative renal perfusion deficits were associated with inflammatory burden and LOS in univariable analyses. CRP showed the most consistent independent association, whereas the incremental value of CT metrics requires external validation.