Background: Artificial intelligence (AI) is increasingly used as a decision-support tool in mammography, but its influence on radiology residents’ interpretive behavior remains insufficiently characterized. This study evaluated the impact of AI assistance on BI-RADS classification and diagnostic confidence among radiology residents. Methods: This retrospective, single-center, multi-reader paired study included 112 diagnostic mammography examinations, corresponding to 223 assessable breasts and 2230 resident-breast readings. Ten radiology residents interpreted 2D mammography and digital breast tomosynthesis examinations first without AI assistance and subsequently with access to AI output. Changes in the BI-RADS category, diagnostic confidence, reasons for modification, and agreement with an expert-consensus BI-RADS reference standard were analyzed. Results: AI-assisted reassessment changed the BI-RADS classification in 9.7% of readings and diagnostic confidence in 19.2%, with any AI-associated modification observed in 24.3% of instances. Upgrades were more frequent than downgrades, particularly for medium- and high-suspicion AI outputs. Confidence increased more often than it decreased. Expert-reference agreement improved modestly, and BI-RADS 4+ sensitivity increased from 72.0% to 82.3%, with stable negative agreement, but these metrics reflect agreement with expert BI-RADS consensus rather than pathology-confirmed cancer detection. Conclusions: AI assistance influenced both BI-RADS reassessment and diagnostic confidence among radiology residents, producing modest but directionally favorable changes. These findings support cautious, supervised integration of AI into breast imaging training, with attention to confidence calibration and potential overreliance.

Background: Magnetic resonance cholangiopancreatography (MRCP) enables non-invasive assessment of biliary lithiasis, duct caliber, and associated anatomical findings. Although age-dependent reference limits for common bile duct (CBD) diameter have been described, the relationships among age, gallbladder stone morphology, ductal anatomy, and CBD stones remain incompletely characterized in consecutive radiology cohorts. This study assessed age-associated MRCP patterns and imaging features associated with CBD stones. Methods: We retrospectively analyzed 242 consecutive MRCP examinations performed for biliary indications. Recorded variables included age, sex, gallbladder and CBD stone characteristics, CBD diameter, intrahepatic bile duct (IHBD) dilatation, CBD tortuosity, cystic duct morphology, periampullary diverticulum, microlithiasis, and associated pancreatobiliary findings. Ordered trend tests, group comparisons, correlation analyses, ROC analysis, and a parsimonious multivariable logistic model were used. Model 2 included six predictor parameters for 65 CBD-stone events (10.8 events per variable) and was internally validated with 2000 bootstrap resamples. Results: The cohort had a median age of 63 years (IQR 49–73), and 135 patients (55.8%) were female. GB stones were present in 211 patients (87.2%), and CBD stones in 65 (26.9%). Across ordered age strata, GB stone prevalence decreased (p = 0.006), whereas CBD stone prevalence increased (p = 0.038); CBD diameter and IHBD severity also increased with age (both p < 0.001). CBD-positive patients had larger CBD diameters than CBD-negative patients (median 10.0 vs. 6.5 mm, p < 0.001). In Model 2, CBD diameter was the strongest independent imaging correlate of CBD stones (OR 1.567 per 1 mm, 95% CI 1.310–1.873; p < 0.001), whereas age was not independently associated after adjustment (p = 0.781). CBD diameter alone achieved an AUC of 0.853; Model 2 achieved an AUC of 0.896 and an optimism-corrected AUC of 0.884. Conclusions: Older age groups showed greater ductal involvement and larger CBD caliber in this consecutive MRCP cohort. CBD diameter was the strongest imaging correlate of CBD stones, while age and CBD diameter provided overlapping information in the adjusted model. These findings describe an observational MRCP pattern and require external validation.

Incidental breast abnormalities may be visible on chest computed tomography (CT), although breast tissue is frequently outside the primary diagnostic focus of these examinations. This proof-of-concept study developed and internally evaluated a bilateral-asymmetry-aware multiple-instance learning framework (BAA-MIL) for case-level discrimination between breast cancer and non-malignant control examinations using preselected breast-containing axial chest CT images. The retrospective single-center case–control dataset included 89 CT examinations from 89 unique patients: 49 histopathologically confirmed breast cancer cases and 40 non-malignant control cases established according to predefined case-level reference criteria. The dataset comprised 304 selected axial CT images, yielding 608 unilateral breast-region crops arranged as 304 bilateral crop pairs. Paired breast crops were processed by a shared convolutional encoder, combined through an explicit side-symmetric asymmetry descriptor, and aggregated across axial images using gated attention-based multiple-instance learning. Performance was assessed using exploratory five-fold stratified patient-level cross-validation and a nested patient-level validation analysis. Under the primary nested patient-level validation protocol, the compact-encoder BAA-MIL implementation achieved an accuracy of 0.528 (95% CI, 0.427–0.629), balanced accuracy of 0.555 (95% CI, 0.465–0.637), sensitivity of 0.286 (95% CI, 0.160–0.415), specificity of 0.825 (95% CI, 0.694–0.930), and ROC AUC of 0.547 (95% CI, 0.420–0.665). The original non-nested analysis yielded higher exploratory estimates, but this analysis used the held-out folds for early stopping and checkpoint selection and is therefore reported only as secondary exploratory analysis. These findings support technical feasibility for a constrained case-level classification task using preselected chest CT images, but do not demonstrate detection of unsuspected incidental breast cancer across complete CT examinations. Performance estimates remain uncertain, and external validation using complete volumetric CT data is required before clinical use.

Automated colorectal tumor segmentation on routine computed tomography (CT) remains technically challenging because of limited soft-tissue contrast, heterogeneous tumor morphology, complex bowel anatomy, and marked imbalance between tumor and background pixels. This retrospective technical feasibility study aimed to develop and evaluate an attention-enhanced ResNet50–U-Net architecture for automated segmentation of primary colorectal tumors on contrast-enhanced abdominopelvic CT images. The dataset comprised 493 axial CT image–mask pairs containing visible colorectal tumors, including 70 institutional images obtained from 25 patients and 423 images from publicly available colorectal cancer imaging resources. Reference masks were generated by manual tumor delineation performed by an experienced radiologist and reviewed by a second abdominal radiologist, with uncertain contours resolved by consensus. The proposed model combined an ImageNet-pretrained ResNet50 encoder with an attention-guided U-Net decoder and was trained using a composite Binary Cross-Entropy and Focal Tversky loss function to address foreground–background class imbalance. Performance was assessed using five-fold image-level cross-validation. Predicted probability maps were binarized using a threshold of 0.5, and segmentation metrics were calculated through global micro-averaging within each validation fold. The model achieved a mean Intersection over Union of 0.7406 ± 0.0276, a Dice similarity coefficient of 0.8507 ± 0.0182, a pixel-level sensitivity of 0.9559 ± 0.0266, and a pixel-level background specificity of 0.9956 ± 0.0003. Qualitative assessment demonstrated substantial spatial agreement between predicted masks and expert annotations, although minor boundary discrepancies and small false-positive components were observed. These findings support the technical feasibility of attention-enhanced encoder–decoder architectures for colorectal tumor segmentation on routine CT images. However, because validation was performed at the image level, entirely tumor-negative images were not included (preventing the evaluation of clinical specificity and false-positive rates), and no independent external test cohort was available, further patient-level, multicenter, and DICOM-based validation is required before clinical implementation. While the model demonstrated robust internal performance, future studies must prioritize large-scale external validation and patient-level cross-validation to rigorously assess generalizability and specificity.

Purpose: To evaluate whether proton MR spectroscopy (¹H-MRS) improves the differentiation of tumoral from non-tumoral intracranial lesions beyond conventional MRI, and to quantify its effect on diagnostic confidence, in a consecutive real-world bicentric cohort of diagnostically ambiguous lesions. Methods: This retrospective, bicentric observational study screened 122 consecutive patients who underwent brain MRI with ¹H-MRS at two imaging centers, of whom 99 were eligible for final analysis after excluding 23 patients due to insufficient follow-up (<12 months) or incomplete reference standard. The reference standard was histopathological confirmation in 43 cases (43.4%) and structured clinico-radiological follow-up of at least 12 months in 56 cases (56.6%). The reporting radiologist’s binary diagnostic impression and confidence score were recorded before and after ¹H-MRS. Conventional MRI alone was compared to MRI plus ¹H-MRS using the exact McNemar test; Cho/NAA discrimination was assessed by ROC analysis with bootstrap confidence intervals, and confidence change by the Wilcoxon signed-rank test. Secondary analyses comprised an intention-to-diagnose analysis of all 99 examinations, counting non-diagnostic spectra as test failures, and an evaluable-case sensitivity analysis restricted to histopathologically confirmed patients with interpretable spectra. Results: Of 99 examinations, 89 (89.9%) yielded interpretable spectra. In the primary analysis, qualitative MRS interpretation demonstrated sensitivity of 76.0% (95% CI: 61.8–86.9), specificity of 94.9% (95% CI: 82.7–99.4), and accuracy of 84.3% (CI: 75.0–91.1), compared with 54.0%, 89.7% and 69.7% (59.0–79.0) for conventional MRI alone (exact McNemar p = 0.024). ROC analysis of the Cho/NAA ratio (n = 89) yielded an AUC of 0.858 (0.774–0.931); the exploratory, cohort-specific Youden-optimal cut-off was 1.41 (sensitivity 78.0%, specificity 87.2%). In the intention-to-diagnose analysis including all 99 examinations, accuracy was 78.8% (69.4–86.4). Diagnostic confidence increased significantly after ¹H-MRS (median 1 to 2; Wilcoxon p < 0.001; effect size r = 0.81), with moderate or major added value in 71 of 99 examinations (71.7%). Conclusions: In a diagnostically heterogeneous real-world cohort, ¹H-MRS significantly improved the accuracy achievable with conventional MRI alone and substantially increased reported diagnostic confidence, with high specificity but only moderate sensitivity. A zone-based Cho/NAA interpretative framework better reflects biological overlap than rigid binary thresholds. As this study assessed diagnostic accuracy and reported confidence rather than therapeutic decisions or patient outcomes, ¹H-MRS should be regarded as an adjunctive decision-support modality rather than a standalone classifier in routine neuro-oncologic workflows.