Insights into Imaging最新文献

Cystic female pelvic lesions, whether of ovarian or non-ovarian origin, are prevalent in routine clinical practice, with the majority originating from gynaecological (ovarian) structures, ranging from functional cysts to malignant ovarian tumours. Despite the fact that we encounter these lesions in the course of our routine clinical work, arriving at an accurate diagnosis can often prove challenging due to the overlap of imaging appearances. Ultrasound is the primary imaging modality for the evaluation of most cystic female pelvic lesions, while MRI serves as a problem-solving tool. In cases that are more complex or equivocal, pelvic MRI proved to be particularly useful due to its superior soft tissue resolution, multiplanar imaging capability and non-invasive nature. In order to make an accurate diagnosis, it is crucial to have a comprehensive understanding of pelvic topographic anatomy, be familiar with possible differential diagnoses and include all relevant clinical data. The classification of ovarian cystic lesions was undertaken using the O-RADS MRI risk stratification system, which provides standardised language for communication between radiologists and clinicians. The objective of this review is to illustrate the spectrum of typical MRI characteristics of different cystic female lesions of both ovarian and non-ovarian origin, with the emphasis on differential diagnoses. The review includes tables with MRI appearances on T2, T1, DWI sequences and postcontrast tomograms. To facilitate the learning process, schematic representations of MRI appearances of ovarian lesions have been incorporated. CRITICAL RELEVANCE STATEMENT: MRI diagnosis of various ovarian and non-ovarian cystic female pelvic lesions and their differential diagnosis. KEY POINTS: The diagnosis of cystic female pelvic lesions can be challenging due to the overlapping imaging characteristics exhibited by these lesions. Discrimination between ovarian and non-ovarian lesions is of paramount importance, given the existence of marked discrepancies in both prognosis and management. If the lesion is of ovarian origin, the O-RADS MRI risk stratification system should be implemented in order to ascertain the risk of malignancy.

Objective: Medical imaging databases suitable for training machine learning/computer vision algorithms are scarce, limiting the potential for development and generalisation of clinical tools. Clinical trial databases are a source of data, known for their high-quality data and reliable annotations. However, they are not tailored to the needs of machine learning or deep learning models. Our objective was to develop a methodology and tools that enable the curation of these databases specifically for the training or testing of artificial intelligence tools.

Materials and methods: MRIs from the French centres of the EURAD clinical trial (MRI of women with pelvic adnexal lesions) were used to constitute the database. We developed the steps required to curate a clinical trial database: definition of inclusion and exclusion criteria, removal of unnecessary data according to the principle of parsimony, quality control, and harmonisation.

Results: A total of 713 patients were included in our study. The directory structure was simplified, and the number of files and folders decreased by 44% and 95% respectively. Only 62 DICOM fields were considered necessary for artificial intelligence (AI) model applications. Quality control was implemented in repeated cycles of automatic checks, followed by a final manual random inspection. Finally, sequence names were harmonised for easy identification when developing models.

Conclusion: Using a clinical trial database, we propose a methodology to build a database suitable to train or test AI algorithms. This study underlines the need for a more global and systematic framework for the secondary use of health data to develop AI imaging tools for patient care.

Critical relevance statement: We propose and detail a framework and tools to curate a clinical trial database to allow secondary use of the high-quality annotated data generated in clinical trials for the training and testing of artificial intelligence models.

Key points: Clinical trial imaging databases are not adapted for AI model development. A curation process of MRI databases was developed for machine learning applications. We share the open-source tools and methodology developed in this study.

Objectives: To evaluate intratumoral subregional and peritumoral radiomics for predicting pathological T stage of clear cell renal cell carcinoma (ccRCC), and investigate the biological mechanisms of radiomics.

Materials and methods: This retrospective study included 323 ccRCC patients from two centers, divided into training (n = 148), internal test (n = 38), and external validation (n = 137) sets. Patients were stratified into low (T1 and T2, n = 222) and high (T3 and T4, n = 101) T stage groups. The tumors were segmented into different intratumoral subregions via the Gaussian mixture model (GMM). Radiomic features (RFs) were extracted from the whole tumor region (VOI_whole), intratumoral subregions (VOI_subx), and the peritumoral region (VOI_peri). Several machine learning (ML) models and radiomic score (Radscore) were developed to predict pathological T stage and prognosis of ccRCC. Radiogenomics analysis was used to explore the relationship between radiomics and biologic pathways.

Results: Two intratumoral subregions were segmented. The support vector machine (SVM)-based combined model, constructed using RFs from VOI_sub1 and VOI_peri, achieved the highest AUC values, of 0.82 (95% CI: 0.68-0.96) and 0.80 (95% CI: 0.71-0.88) in the internal test and external validation sets, respectively. A higher Radscore was correlated with poorer overall survival (OS) (p < 0.001). Radiogenomics analysis revealed that radiomics was associated with extracellular matrix remodeling, vesicle transport, protein processing in the endoplasmic reticulum, and the Hippo signaling pathway.

Conclusions: An ML model combining intratumoral subregion and peritumoral RFs showed good performance in predicting the pathological T stage of ccRCC, and these RFs were associated with biological pathways underlying tumor invasion.

Critical relevance statement: This study develops a validated CT-radiomics model (intratumoral subregions + peritumoral) predicting ccRCC T stage. The prognostic Radscore links to invasion biology (ECM remodeling, Hippo/ER dysregulation), enabling clinical translation.

Key points: Subregional and peritumoral radiomics models accurately predicted ccRCC (clear cell renal cell carcinoma) histological T stage. Radiomics score identified that high-risk ccRCC patients had poorer overall survival. Predictive radiomic features (RFs) were associated with biological pathways underlying tumor invasion.

Patients with end-stage renal disease (ESRD) develop brain iron deposition due to iron metabolism disorders induced by long-term hemodialysis. This abnormal iron accumulation accelerates cognitive impairment (CI) and neurodegenerative pathologies. Quantitative susceptibility mapping (QSM), a technique capable of precisely quantifying magnetic susceptibility, provides a novel perspective for the noninvasive and dynamic monitoring of cerebral iron distribution. Monitoring brain iron deposition using QSM facilitates the development of individualized clinical treatment strategies. This review systematically examines the application of QSM in studying brain iron deposition in hemodialysis patients, with a focus on analyzing the dynamic patterns of iron deposition pre- and post-dialysis and during follow-up periods. It further explores the relationship between QSM findings and iron metabolism dysregulation, blood-brain barrier (BBB) injury, and oxidative stress. Additionally, the predictive value of QSM for clinical neurological functional prognosis following iron chelation therapy is discussed. CRITICAL RELEVANCE STATEMENT: QSM studies on cerebral iron deposition in hemodialysis patients require further monitoring of its spatial-temporal dynamics and changes after iron chelation. Future research should focus on technical standardization, longitudinal tracking, and treatment response to establish a precision neuroimaging-guided framework. KEY POINTS: This review exploration is warranted to monitor the spatial distribution and dynamic changes of brain iron deposition in this population. The relationships between QSM findings and iron metabolism dysregulation, blood-brain barrier injury, and oxidative stress are explored. This review focuses on issues in the fields of technology standardization, longitudinal monitoring, and treatment responsiveness.

Objectives: Artificial intelligence (AI) could facilitate and objectify quality assessment in the daily routine. The purpose was to explore the extent to which an AI prototype algorithm is able to replicate the perfect-good-moderate-inadequate (PGMI) system (perfect, good, moderate, inadequate).

Materials and methods: From a multicentre case collection, 200 standard mammograms (800 images) were selected. A deep learning-based prototype software was used to rate the images in analogy to the PGMI system. The AI results were compared with a reference standard obtained through consensus reading by three expert radiographers and one expert radiologist, using quadratically weighted Cohen's kappa with confidence intervals (CI) and context-based interpretation. Frequency and reasons for disagreement were evaluated for challenging cases with a discrepancy of two or more grades and a discrepancy in assigning an inadequate.

Results: For overall PGMI per image, slight agreement between human consensus and AI was observed for CC views (κ = 0.14) and fair agreement for MLO views (κ = 0.25). The highest agreement was observed for the CC category "M. Pectoralis visibility" (substantial, κ = 0.75). Best category in MLO was "Pectoralis angle" (moderate, κ = 0.49). For other categories, fair, slight or poor agreement was observed. The work-up of disagreement gave insight into misinterpretations of anatomical landmarks and causality issues in the categorization.

Conclusion: Transforming the PGMI system into a fully automated AI algorithm is challenging and may differ substantially between subcategories. Further research in computer science and quality assessment methodology is needed to pave the way for AI-based objective quality management in mammography.

Critical relevance statement: Profound evaluation of AI algorithms and their ability to replicate human interpretation, scoring, and classification are the basis and scientific framework toward AI-based objective quality management in mammography.

Key points: AI has huge potential for automated assessment of diagnostic image quality. Compared with human reading agreement, substantial disagreement may also be found. Direct transformation of perfect-good-moderate-inadequate scoring into an AI algorithm is challenging.

Objectives: To test the feasibility of 60 kVp double-low-dose coronary CT angiography (CCTA) with a deep learning reconstruction (DLR) algorithm.

Materials and methods: Eighty-nine patients (44 females, 59.9 ± 13.2 years, 23.1 ± 3.3 kg/m²) with known or suspected coronary artery disease were prospectively enrolled. Each patient underwent the double-low-dose CCTA (60-kVp, 28 mL contrast at 2.5 mL/s) and was immediately followed by routine-dose CCTA (100-kVp, 44 mL contrast at 4.0 mL/s). Routine-dose data were reconstructed using hybrid iterative reconstruction (RD-HIR), and double-low-dose data were reconstructed using both HIR (LD-HIR) and DLR (LD-DLR). The consistency of both coronary stenosis assessments and CT-derived fractional flow reserve (CT-FFR) values between low-dose and routine-dose images was quantified using receiver operating characteristic analysis at various levels. Segment-level image quality scores (IQS), signal-noise-ratio (SNR), and contrast-noise-ratio (CNR) were compared among three groups.

Results: Double-low-dose CCTA achieved a significant reduction in both radiation dose (0.60 ± 0.12 mSv vs 4.43 ± 1.42 mSv) and contrast volume compared to routine-dose CCTA. For the per-segment level, LD-DLR showed significantly higher specificity (0.99 vs 0.94), positive predictive value (0.91 vs 0.68), and accuracy (0.98 vs 0.94) for ≥ 50% coronary stenosis compared to LD-HIR. The area under the curve of LD-DLR was significantly higher than LD-HIR for ≥ 50% stenosis at per-segment (0.94 vs 0.92), per-vessel (0.92 vs 0.89), and per-patient (0.92 vs 0.85) levels; and for CT-FFR ≤ 0.80 at per-vessel (0.94 vs 0.74), LAD-vessel (0.94 vs 0.71), and LCX-vessel (0.99 vs 0.67) levels. The IQS, SNR, and CNR of LD-DLR were not inferior to those of RD-HIR in all segments.

Conclusions: The 60 kVp double-low-dose CCTA with DLR can significantly reduce radiation dose and simultaneously maintain the high consistency of coronary stenosis and CT-FFR assessments with routine-dose CCTA.

Critical relevance statement: The 60 kVp double-low-dose CCTA protocol is feasible with a novel DLR algorithm without compromising the performance of coronary stenosis and CT-FFR assessments.

Key points: Is a 60 kVp double-low-dose CCTA protocol with a DLR algorithm feasible for routine clinical application? The 60 kVp CCTA protocol with the DLR algorithm reduced radiation dose by 86.5% and contrast dose by 36.4%. The 60 kVp CCTA with DLR achieved high consistency of coronary stenosis and CT-FFR values with the routine-dose 100 kVp CCTA.

Objectives: The study aimed to assess the predictive performance of transition zone PSA density (TZ-PSAD) compared to conventional PSA density (PSAD) in detecting clinically significant prostate cancer (csPCa) among patients with negative pre-biopsy MRI findings.

Materials and methods: The study included 606 patients with negative MRI findings who subsequently underwent transrectal ultrasound-guided systematic biopsy. AI software automatically measured prostate and zonal volumes, from which PSAD and TZ-PSAD (total PSA/transition zone volume) were calculated. Diagnostic performances were evaluated using ROC curve analysis, risk stratification was applied to select patients needing biopsy, and independent predictors of imaging-invisible csPCa were determined through univariate and multivariate analyses.

Results: 51 patients (8.4%) were diagnosed with csPCa. TZ-PSAD demonstrated significant superior discriminative ability (AUC = 0.718) compared to PSAD (AUC = 0.686; p = 0.019). Patients with TZ-PSAD ≥ 0.35 ng/mL/cc had a csPCa detection rate of 20.1%, while those below this threshold had a rate of 4.1%. The optimal TZ-PSAD threshold of 0.35 ng/mL/cc showed superior performance than the guideline-recommended PSAD threshold of 0.2 ng/mL/cc. Multivariate analysis identified TZ-PSAD as a strong independent predictor of imaging-invisible csPCa.

Conclusions: TZ-PSAD outperforms conventional PSAD in predicting csPCa among men with negative MRI, offering a valuable tool for risk stratification. This facilitates individualized risk assessment, potentially reducing unnecessary biopsies and optimizing patient management.

Critical relevance statement: Our AI system delivers accurate and reproducible prostate zone segmentation, while TZ-PSAD derived from AI outperforms conventional PSAD in detecting csPCa in MRI-negative patients and serves as an effective triage tool to optimize biopsy decision-making and reduce unnecessary systematic biopsies.

Key points: Our AI system enables accurate and reproducible segmentation and measurement of prostate zones. TZ-PSAD demonstrates significantly superior diagnostic performance over conventional PSAD for identifying men with a negative MRI who will have csPCa on a systematic biopsy. TZ-PSAD represents an effective triage metric to reduce unwarranted systematic biopsies in MRI-negative patients.

Objectives: To evaluate the impact of a law amendment that reduced the eye lens dose limit on the use of personal dosimeters among radiation workers in medical settings.

Materials and methods: A repeated cross-sectional survey was conducted at medical institutions across three periods: before the law amendment (control) and during the promulgation and implementation periods. Surveyors (radiological technologists) at each participating medical institution recorded dosimeter-wearing status among radiation workers. Data were collected via mail or email and analysed. The observed workers included physicians, nurses, and radiological technologists.

Results: The surveys were collected from 1194 workers in the control period, 1374 in the promulgation period, and 1194 in the implementation period, totalling 3762 workers. Post-law amendment, the overall wearing rate of primary personal dosimeters significantly increased from 64.6% to 77.9% (p < 0.001). Significant increases in wearing rates were observed among physicians and radiological technologists (p < 0.001). Among occupations, physicians showed the lowest wearing rates across all periods (control: 35.8%, promulgation: 56.7%, implementation: 62.6%), whereas radiological technologists showed the highest (control: 92.7%, promulgation: 98.5%, implementation: 99.5%). Regarding physician specialities, orthopaedic surgery exhibited the lowest compliance (control: 11.3%, promulgation: 35.4%, implementation: 24.7%). The proportion of workers without provision of a personal dosimeter declined from 5.9% to 1.9% (p < 0.001).

Conclusions: Despite overall improvement following the law amendment, low compliance among physicians, particularly in orthopaedics, indicates the need for targeted interventions.

Critical relevance statement: Although dosimeter-wearing rates improved after Japan's eye dose limit revision, persistent low physician compliance-especially in orthopaedics-highlights the need for targeted strategies to strengthen radiation protection in clinical practice.

Key points: The effect of reduced eye dose limits on dosimeter use remains unclear. Personal dosimeter usage increased significantly after the law amendment. Compliance remained low among orthopaedic physicians despite regulatory tightening. Targeted interventions are needed for low-compliance groups to ensure radiation protection.

Objectives: To develop and validate a primary tumor-derived, multiparametric MRI-based deep learning-radiomics-clinical (DLRC) model for predicting pelvic lymph node metastasis (LNM) in early-stage cervical cancer.

Materials and methods: This retrospective five-center study selected 1095 patients (Jan 2020-Dec 2022), divided into training (n = 481), internal validation (n = 204), and external validation (n = 410) cohorts. Radiomics and deep learning (DL) features were extracted from the volumetric segmentations of the primary cervical tumors on three MRI sequences (CE-T1WI, DWI, FS-T2WI). After constructing individual radiomics and DL models, the DLRC model was developed by integrating the radiomics_score, optimal DL model, and significant clinical features. Model performance was evaluated using ROC analysis, calibration curves, and decision curve analysis.

Results: The DLRC model demonstrated superior predictive performance, achieving AUCs of 0.807 (95% CI: 0.766-0.849) in the training cohort, 0.789 (95% CI: 0.721-0.857) in the internal validation cohort, and 0.807 (95% CI: 0.761-0.853) in the external validation cohort. It significantly outperformed both the radiomics model and the optimal DL model (all p < 0.001) in the external validation cohort. The calibration curves indicated good agreement between predictions and observations. The decision curve analysis showed that the DLRC model provided the highest net clinical benefit across most decision thresholds.

Conclusions: The DLRC model, which integrates tumor-derived multiparametric MRI features with clinical features, represents a robust and generalizable tool for the preoperative prediction of LNM. Its comparable accuracy to standardized radiological assessment and clinical utility shows potential to aid in personalized treatment planning for patients with early-stage cervical cancer.

Critical relevance statement: The combined model (DLRC) integrating deep learning and radiomics features from the primary tumor with clinical characteristics enables preoperative LNM risk stratification, supporting personalized surgical planning and reducing unnecessary lymphadenectomy.

Key points: Accurate preoperative prediction of lymph node metastasis in early-stage cervical cancer remains a significant clinical challenge. The model integrating deep learning and radiomics features derived from the primary tumor with clinical features achieved robust and generalizable predictive performance. The accuracy of a deep learning-radiomics-clinical nomogram for lymph node metastasis risk stratification in early-stage cervical cancer is comparable to standardized radiological assessment.

Colorectal cancer (CRC) is the third most common malignancy worldwide, and early detection is vital to prevent metastasis and postoperative recurrence. This review summarizes current applications of spectral computed tomography (CT) in CRC, including its principles, spectral parameters used for evaluating primary and metastatic lesions, and key findings from recent literature. A systematic search of PubMed, Web of Science, and Google Scholar identified English-language studies published between April 2018 and April 2025 using the keywords: "spectral CT," "spectral imaging," "dual-layer spectral CT," "dual-energy spectral CT," "colorectal cancer," and "colon cancer." Spectral CT has shown promise in improving CRC detection and T staging accuracy, increasing sensitivity for lesion characterization, and aiding prognostic assessment after chemotherapy using baseline spectral parameters. Early evidence suggests it may also help predict lymph node metastasis and identify patients at risk of early postoperative metastases or surgical complications. Spectral parameters have been correlated with KRAS mutation, Ki-67 index, microsatellite instability, lymphovascular, perineural, and extramural vascular invasion, as well as microvessel density. However, most studies remain small and observational, highlighting the need for validation in larger, multicenter cohorts. Standardization and the time-intensive nature of image segmentation currently limit widespread adoption. Nevertheless, spectral CT is expected to play an increasing role in CRC evaluation by providing quantitative, predictive imaging biomarkers. Integration with artificial intelligence, particularly deep learning and automated segmentation, will likely expand both research and clinical applications. CRITICAL RELEVANCE STATEMENT: This article explores the current applications of spectral CT in colorectal cancer by outlining the fundamentals of spectral CT, the spectral parameters used to assess, stage, and predict the prognosis of primary and metastatic disease, as well as the main findings from the current literature. KEY POINTS: Spectral CT may be helpful in the detection of colorectal primary tumors, lymph node metastases, and liver metastases, as well as in predicting treatment response. Spectral CT offers a non-invasive method to assess genetic mutations and prognostic factors associated with colorectal primaries. The lack of standardization in technology and measurement methods limits its applicability in clinical practice.