Insights Into Imaging

🤖 Artificial intelligence is only as reliable as the reference standard behind it.

🩻 United States National Institute for Occupational Safety and Health-certified B readers demonstrated strong agreement and high accuracy for tuberculosis detection on chest radiographs.

📊 These findings provide a robust benchmark for external validation of artificial intelligence tools in tuberculosis diagnosis.

https://link.springer.com/article/10.1186/s13244-026-02334-0 (Wiwatana Tanomkiat et al.)

Magnetic resonance imaging has transformed prostate cancer staging, but the tumour-node-metastasis classification still relies on digital re**al examination.

📊 Digital re**al examination is inferior for detecting extraprostatic extension
⚖️ Parallel reporting of magnetic resonance imaging-based and clinical T-staging may improve risk stratification
📝 A framework is proposed for future tumour-node-metastasis updates

https://link.springer.com/article/10.1186/s13244-026-02346-w (Georgios Agrotis et al.)

🎓 Objective structured clinical examination data (2019–2024) support its use for assessing nuclear medicine resident competence.

📊 Stations showed acceptable difficulty and good discrimination across clinical tasks
🧪 Reliability was moderate and variable across years
🏥 Effective simulation of clinical practice in imaging, reporting, and patient care

https://link.springer.com/article/10.1186/s13244-026-02340-2 (Lijuan Di et al.)

📊 Interobserver variability remains a key limitation in RECIST tumor response assessment.

🔍 Variability is driven by lesion selection, measurement methods, imaging quality, and reader experience
🧑‍⚕️ Inconsistencies in statistical approaches also reduce comparability across studies
📚 Standardized protocols are needed to improve reproducibility in oncology trials

https://link.springer.com/article/10.1186/s13244-026-02320-6 (Illaa Smesseim et al.)

Contrast-enhanced mammography is gaining interest for preoperative breast cancer staging.

📊 Tumor extent assessment and detection of additional lesions are comparable to breast magnetic resonance imaging in selected settings
🎯 Influences surgical planning in ~20–30% of patients
⚠️ Limitations include radiation dose, posterior structure evaluation, and heterogeneous protocols

https://link.springer.com/article/10.1186/s13244-026-02331-3 (Chiara Bellini et al.)

An MRI-derived intratumoral heterogeneity score (ITHscore) was developed to estimate 21-gene recurrence risk in ER+/HER2− breast cancer.

📊 Higher ITHscore was associated with high-risk recurrence groups
🧠 Adding ITHscore improved prediction performance (AUC up to 0.86)
📈 Model showed good generalization in an external cohort

https://link.springer.com/article/10.1186/s13244-026-02342-0 (Yang Chen et al.)

🎓 A competency-based modular training model was evaluated in radiology residency education.

📚 Interdisciplinary, disease-focused modules integrated with clinical rotations and image post-processing
📊 No overall score difference vs traditional training
👩‍⚕️ Significant improvement in non-graduate subgroup (theory and practice)

https://link.springer.com/article/10.1186/s13244-026-02335-z (Shang Wan et al.)

Ultrasound radiomics combined with clinical features improves breast lesion diagnosis in a large multicenter study.

📊 Combined model outperformed radiomics and clinical models alone (AUC up to 0.93)
📉 Potential to reduce unnecessary biopsies while maintaining diagnostic sensitivity
🧠 Stable performance across external and prospective validation cohorts

https://link.springer.com/article/10.1186/s13244-026-02344-y (Di Zhang et al.)

A multimodal CT urography model improves perioperative risk stratification in upper tract urothelial carcinoma.

📊 Fusion of clinical data, radiomics, and deep learning outperforms single-modality models
📈 Highest predictive performance achieved for 3-year overall survival (external AUC ~0.77)
🧠 2.5D deep learning adds incremental prognostic value over 2D approaches

https://link.springer.com/article/10.1186/s13244-026-02337-x (Xiang Peng et al.)

🖐️ Foveal triangular fibrocartilage complex lesions are frequently underreported on wrist arthro-computed tomography.

📊 Review identified lesions in 59% of cases, with many ulnar/foveal injuries missed in initial reports
🔍 Structured re-evaluation improved detection in 28% of cases and aligned better with surgical findings
⚠️ Foveal involvement was strongly associated with ulnar-sided wrist pain

https://link.springer.com/article/10.1186/s13244-026-02327-z (Julien Dejean-Servieres et al.)