Background: Menie`re's disease (MD) is a heterogeneous inner ear disorder, and imaging-based endotype identification may inform prognosis and personalized management. The vestibular aqueduct (VA) is a candidate imaging biomarker, and photon-counting CT (PCCT) offers high-resolution depiction of temporal bone microanatomy. This study evaluated PCCT-derived VA morphometric parameters as imaging biomarkers for MD endotype identification and hearing loss assessment. Methods: In this retrospective study, 106 patients (212 ears) underwent temporal bone PCCT and 3.0-T 3D-FLAIR MRI. PCCT parameters included angular trajectory of the VA (ATVA), VA visibility grade, morphologic subtype, midpoint width, VA length, and posterior semicircular canal-posterior fossa distance. Two neuroradiologists independently evaluated all images. Diagnostic performance for six clinically motivated tasks was assessed by AUC with leave-one-patient-out cross-validation (LOPOCV); associations with pure-tone average (PTA) were evaluated by multivariable linear regression with cluster-robust standard errors. Results: Interobserver agreement was excellent (ICC, 0.84-0.91). VA visibility grading identified the hypoplastic endotype with an AUC of 0.860 (LOPOCV, 0.812); combined VA morphology and midpoint width distinguished the degenerated endotype with an AUC of 0.784 (LOPOCV, 0.736). Multivariable regression identified ATVA as an independent PTA predictor (beta = 0.238, p = 0.023). Conclusions: PCCT-derived VA morphometric parameters enabled MD endotype identification despite clinical heterogeneity and were associated with hearing loss severity, supporting their potential role as imaging biomarkers for MD management.
Background PI-RADS scoring of transition zone (TZ) lesions is strongly reader-dependent, particularly in the PI-RADS 4 category. We compared diagnostic performance and inter-reader agreement between a specialized high-volume center and non-specialized referring centers. Methods Overall, 315 men with a TZ lesion scored PI-RADS ≥ 3 underwent MRI-guided in-bore biopsy and were stratified by interpreting site (non-specialized vs specialized). The erased-margins criterion was reassessed centrally by one specialized-center reader (the biopsy operator) and, in a 125-lesion subset, by three independent readers. Results Positive predictive value for clinically significant cancer (csPCa, Grade Group ≥ 2) at PI-RADS ≥ 4 was higher in the specialized cohort (43.3% vs 29.5%, p = 0.015). For PI-RADS 4 lesions, detection of any cancer was 76.9% vs 42.6% (p < 0.001), but not significantly different for csPCa (32.1% vs 22.8%, p = 0.16). On central reassessment, non-specialized centers had more often assigned PI-RADS 4 to lesions without erased margins (50% vs 35%, p = 0.048), a difference not replicated by the three independent readers. Agreement on PI-RADS category was substantial within and moderate between centers (κ 0.68 vs 0.43–0.49); agreement on the erased-margins criterion was fair-to-moderate (Fleiss κ 0.30), and its discriminative value decreased with decreasing expertise (AUC 0.82, 0.75, 0.70), largely reflecting lesion identification. The central reassessment did not outperform blinded readers (AUC 0.69). Conclusion Specialized interpretation was associated with higher diagnostic yield for TZ lesions, particularly PI-RADS 4. Reader-dependent application of the erased-margins criterion is a candidate mechanism requiring confirmation and a potential training target rather than a stand-alone rule; prospective external validation is needed.
Purpose Large language models (LLMs) are increasingly explored for medical applications. This study evaluates the agreement of three LLMs—ChatGPT, Copilot, and Gemini—with two expert radiologists in assigning ACR TI-RADS categories to thyroid nodules, using fine-needle aspiration cytology as the reference standard. Methods A retrospective analysis included 135 thyroid nodules from 135 unique patients (108 benign, 27 malignant). Structured TI-RADS feature scores and final categories (TR1-TR5) were extracted from existing reports generated by each LLM and two board-certified radiologists (≥8 years’ experience). Agreement was assessed using Cohen’s κ (feature-level), quadratic weighted κ (category-level), and diagnostic performance (sensitivity, specificity, area under the ROC curve [AUC]) with TR4-TR5 classified as test-positive categories for the binary diagnostic analysis. McNemar’s test evaluated discordant classification patterns. Results Feature-level agreement was fair to substantial for composition and margin (κ up to 0.621) but poor for echogenic foci (κ as low as 0.075). Weighted κ for TI RADS categories was moderate to substantial for ChatGPT (κw 0.614–0.621) and Gemini (0.516–0.621), but fair for Copilot (0.383–0.314). All LLMs showed high sensitivity (81.5–96.3 %) but low specificity (20.4–38.0 %). Ordinal AUC values were 0.757 (ChatGPT), 0.644 (Copilot), and 0.619 (Gemini). McNemar’s test suggested a more balanced discordant pattern for ChatGPT than for Copilot and Gemini. Conclusion ChatGPT, Copilot, and Gemini show moderate composition/margin agreement but poor echogenic foci detection and overclassify benign nodules; standalone use is not recommended. Domain‑specific fine‑tuning and prospective validation are required.
Background Positron emission tomography (PET) enables non-invasive whole-body assessment of molecular biomarkers and is particularly valuable for evaluating heterogeneous diseases such as cancer. Platelet-derived growth factor receptor beta (PDGFRβ) is a stromal biomarker expressed on activated pericytes and fibroblasts that contributes to tumour remodelling, vascular regulation and therapeutic resistance. This study describes the translational implementation of the PDGFRβ-targeting PET tracer [68Ga]Ga-ATH001, from GMP-compliant radiopharmaceutical development to its first reported oncology application in patients with liver tumours. Methods A GMP-compliant automated synthesis of [68Ga]Ga-ATH001 was established using a ModularLab PharmTracer synthesizer and validated for clinical use. The tracer was prospectively implemented in the IMAGINE-1 study (EU CT 2025-522290-11-00). PET imaging included dynamic liver acquisition immediately after injection, followed by whole-body PET/CT imaging and delayed liver scans to assess tracer biodistribution, tumour uptake and tracer kinetics. Results [68Ga]Ga-ATH001 was successfully translated into clinical production using an automated in-house synthesis process, achieving a decay-corrected radiochemical yield of 51.9 ± 0.7 %, radiochemical purity of 97 %, radiochemical stability of 96.3 ± 0.5 % after 2 h at room temperature and apparent molar activity of 47.2 ± 1.6 GBq/µmol at EOS. Three consecutive production runs assessed reproducibility; a fourth batch served as a microbiological worst-case challenge and was excluded from the assessment. PET/CT imaging in three patients with colorectal liver metastases showed SUVpeak tumour-to-background ratios of 1.5 to 1.8 at 60 min after injection. Conclusion GMP-compliant production enabled the first reported oncology application of [68Ga]Ga-ATH001 in patients with liver tumours. The ongoing phase 1 IMAGINE-1 study will further evaluate its clinical potential. EU trial registration Clinical Trials Information system: CTIS, EU CT 2025-522290-11-00, Registered 29 September 2025, IMAGINE-1 | 2025-522290-11-00 | CTIS.eu
