To improve the accuracy of machine learning models for preoperative prediction of high-intensity focused ultrasound (HIFU) ablation efficacy for uterine fibroids by correcting class imbalance in small sample datasets using undersampling methods. Clinical and imaging data were collected from 140 patients with uterine fibroids undergoing HIFU treatment at Foshan Women and Children Hospital, including 104 with high ablation rates and 36 with low ablation rates. Radiomic features were extracted from MRI T2-weighted images (T2WI) of the patients, and machine learning models were constructed to predict HIFU treatment outcomes. Four machine learning algorithms, including k-Nearest Neighbors (KNN), Random Forest (RF), Support Vector Machine (SVM), and Multilayer Perceptron (MLP), were coupled with 7 undersampling methods, namely Random Undersampling (RUS), Repeated Edited Nearest Neighbors (RENN), All k-Nearest Neighbors (AllKNN), Neighborhood Cleaning Rule-3 (NM), Condensed Nearest Neighbor (CNN), Neighborhood Cleaning Rule (NCR), and Instance Hardness Threshold (IHT), for handling class imbalance in the datasets. The 28 prediction models were evaluated using 5-fold cross-validation for areas under the receiver operating characteristic curve (AUC), accuracy, recall, and specificity. The best combinations of undersampling methods and machine learning models CNN-RF, NM-SVM, CNN-KNN, and NM-MLP had AUCs of 0.772 (95% <i>CI</i>: 0.566-0.942), 0.797 (95% <i>CI</i>: 0.600-0.950), 0.822 (95% <i>CI</i>: 0.635-0.964), and 0.822 (95% <i>CI</i>: 0.632-0.960), respectively. The AUCs of the machine learning models significantly increased after coupling with undersampling methods, with the MLP model showing the most pronounced improvement. The recall rates of the 4 combined models also improved significantly (by 0.389 for CNN-RF, 0.836 for NM-SVM, 0.532 for CNN-KNN, and 0.372 for NM-MLP). The use of undersampling methods can effectively correct class imbalance in small sample datasets to improve the accuracy of machine learning models for predicting the efficacy of HIFU ablation for uterine fibroids.

Large language models (LLMs) are emerging artificial intelligence technologies with strong text and image processing capabilities, offering critical support for the intelligent transformation of healthcare and improving clinical efficiency and quality. This review summarizes the current applications, technical features, and future directions of LLMs in cancer diagnosis, focusing on two key scenarios: automated analysis of textual reports (e.g., imaging, pathology, and case summaries) and multimodal diagnosis combining text and medical images. Findings show that LLMs now perform at a level comparable to general resident physicians in cancer diagnosis but are still incapable of making specialized and precise judgments. They also exhibit application-specific traits, such as parameter-efficient models adapted for grassroots-level scenario and divergent versatility in multilingual report analysis. Future efforts should prioritize developing specialized, practical medical LLMs through optimized fine-tuning strategies, construction of high-quality Chinese medical datasets, and integration with vision-language models to promote the clinical application of these models and increase the accessibility of healthcare resources.

Mammographic density is associated with the risk of developing breast cancer and can be predicted using deep learning methods. Model uncertainty estimates are not produced by standard regression approaches but would be valuable for clinical and research purposes. Our objective is to produce deep learning models with in-built uncertainty estimates without degrading predictive performance.&#xD;&#xD;Approach: We analysed data from over 150,000 mammogram images with associated continuous density scores from expert readers in the Predicting Risk Of&#xD;Cancer At Screening (PROCAS) study. We re-designated the continuous density scores to 100 density classes then trained classification and ordinal deep learning models. Distributions and distribution-free methods were applied to extract predictions and uncertainties. A deep learning regression model was trained on the continuous density scores to act as a direct comparison.&#xD;&#xD;Results: The root mean squared error (RMSE) between expert assigned density labels and predictions of the standard regression model were 8.42 (8.34-8.51) while&#xD;the RMSE for the classification and ordinal classification were 8.37 (8.28-8.46) and 8.44 (8.35-8.53) respectively. The average uncertainties produced by the models were higher when the density scores from pairs of expert readers density scores differ more, when different mammogram views of the same views are more variable, and when two separately trained models show higher variation.&#xD;&#xD;Conclusions: Using either a classification or ordinal approach we can produce model uncertainty estimates without loss of predictive performance.

The role of adipose tissue in predicting microvascular invasion (MVI) in patients with hepatocellular carcinoma (HCC) remains unclear. This study proposes a method that integrates deep learning and machine learning techniques to investigate the role of adipose tissue in identifying MVI status in HCC patients. We collected enhanced Computed Tomography images from 517 HCC patients across two independent centers, dividing them into a training set, validation set, and test set. The model was constructed using adipose and tumor deep learning features along with clinical features, and the features were input into a classifier for prediction. The model performance was evaluated using the area under the curve(AUC), decision curve analysis, scatter plots, and box plots. Furthermore, we compared the model's performance with that of three radiologists. After incorporating the adipose tissue modality, the venous-phase AUC reached 0.866 (95% CI 0.803-0.920), while the arterial-phase AUC was 0.864 (95% CI 0.792-0.920). The inclusion of the adipose tissue modality provided significant value for clinical diagnosis, which was further validated through visualization analysis. Using predicted labels for grouping, it shows that the overall survival of the high-risk group was significantly lower than that of the low-risk group. Comparative analysis showed that the predictive performance of the model surpassed that of radiologists. Univariate analysis identified the adipose region as a risk factor for predicting MVI status. We developed a hybrid multimodal model that performed comparably to radiologists' assessments. The inclusion of the adipose tissue modality enhanced the accuracy of MVI diagnosis.

This study aims to both develop and evaluate a predictive model for ureteral access sheath(UAS)placement success using preoperative CT-based 3D ureteral imaging and machine learning techniques. Specifically, it investigates the impact of ureteral anatomical angles on UAS placement success and integrates these angles with multiple machine learning models for preoperative risk stratification. The study also assesses the performance of these models, providing insights into their predictive accuracy and clinical applicability. We retrospectively analyzed 302 patients who underwent initial flexible ureteroscopy lithotripsy (FURS) from January 2022 to August 2023 at Xiangya Hospital, Zhuzhou. None had preoperative ureteral stents. Preoperative CT scans were used to reconstruct the lower ureter in 3D and measure key anatomical angles. Logistic regression identified independent predictors of UAS placement success. Eight machine learning models were developed, with SHAP analysis applied to assess each variable's contribution to prediction accuracy. The UAS placement success rate was 71.19%. Univariate analysis found that both the angle between the ureteral orifice and body axis (∠α; OR = 0.94, 95% CI: 0.89-0.99, p = 0.019) and the angle between the outermost segment of the lower ureter and body axis (∠β; OR = 0.93, 95% CI: 0.89-0.97, p < 0.001) were significantly associated with success. Multivariate analysis confirmed ∠β as an independent predictor (OR = 0.95, 95% CI: 0.90-0.99, p = 0.024). SHAP analysis highlighted ∠β as the most influential variable, with failure risk rising sharply when ∠β exceeded 40°. The ∠β is a critical independent factor affecting UAS placement success. Integrating 3D CT measurements with machine learning allows quantitative risk assessment, aiding in preoperative planning and personalized surgical decision-making. This approach shows strong potential for clinical application.

Radiation dose reduction is essential in paediatric lung computed tomography (CT). Advances in energy-integrating detector CT and deep-learning reconstruction may enable ultra-low-dose imaging comparable to photon-counting CT. To evaluate the radiation dose and performance of an ultra-low-dose lung CT protocol using a wide-detector energy-integrating CT system in paediatric patients, focusing on effective radiation dose and diagnostic image quality. A total of 277 low-dose lung CT scans from 106 paediatric patients (age range, 113 days to 17.85 years) were retrospectively analysed. All scans were acquired in axial mode using a 256-slice-multidetector CT scanner with deep learning image reconstruction and attenuation-based Auto Prescription. Radiation dose parameters, including volume CT dose index, dose-length product, size-specific dose estimate, and effective dose, were calculated. Signal-to-noise ratio and contrast-to-noise ratio were assessed in standardised anatomical regions. Patients were stratified by age, and statistical analysis was conducted to evaluate dose trends and image quality metrics. There were significant differences between all age groups for all dose parameters (Kruskal-Wallis test, P<0.05). The median effective dose increased with age, ranging from 0.12 mSv (interquartile range (IQR) 0.09-0.14 mSv) in the 0-5-year group to 0.23 mSv (IQR 0.21-0.25 mSv) in adolescents aged 15 years to <18 years. Contrast-to-noise ratio and signal-to-noise ratio exhibited age-dependent variation with a small increase in older age groups. One-sided non-inferiority testing demonstrated that the signal-to-noise ratio and contrast-to-noise ratio in the youngest age group (0-5 years) were not significantly inferior to those in the ≥15-year group (P<0.05). All examinations were deemed diagnostically sufficient by board-certified paediatric radiologists. Non-disruptive artefacts such as cardiac motion and step artefacts occurred frequently but did not impair interpretation. Ultra-low-dose lung CT using wide-detector energy-integrating CT with deep-learning image reconstruction allows for routine diagnostic imaging in children at radiation doses ranging from 0.12 mSv to 0.23 mSv, comparable to those reported for newer photon-counting CT systems. This approach provides a robust, clinically viable strategy for minimizing radiation exposure while maintaining diagnostic image quality.

Sarcopenia, characterized by progressive skeletal muscle loss, is associated with poor outcomes in various diseases. Traditional methods for assessing muscle cross-sectional area using computed tomography (CT) scans are manual, time-consuming and prone to variability. This study comprehensively validates a deep-learning (DL) pipeline for accurate and reproducible sarcopenia detection on computed tomography across diverse disease abdominal conditions and imaging protocols. We utilized the publicly available Sparsely Annotated Region and Organ Segmentation (SAROS) CT dataset (n = 550 CT scans, 6516 slices) for model training. Testing was conducted on 601 CT scans from public (SAROS, Cancer Imaging Archive [TCIA] , WAW-TACE) and in-house multi-center datasets representing varied clinical conditions (acute pancreatitis, inflammatory bowel disease, gallbladder cancer and distal bile duct obstruction). The implemented pipeline integrated TotalSegmentator for L3 vertebral segmentation, automated L3 slice extraction and skeletal muscle segmentation using nnU-Net. Performance evaluation included expert qualitative scoring, Dice scores, intersection over union (IoU) and diagnostic accuracy metrics for sarcopenia detection. The DL pipeline demonstrated consistent segmentation accuracy across diverse datasets, with mean Dice scores ranging from 0.9287 to 0.9701 and mean IoU values up to 0.9423. Expert evaluation confirmed reliable L3 vertebral segmentation (78%-85% rated as complete) and skeletal muscle segmentation (90%-92.6% rated as excellent). Sarcopenia detection was consistent across varied patient populations, with sensitivity (0.94-0.97), specificity (0.84-0.97) and AUC values up to 0.92. Importantly, sub-group analysis confirmed comparable performance across varying disease conditions, CT protocols, contrast usage and radiation doses. This study demonstrates that a deep-learning pipeline can achieve consistent and reliable performance for skeletal muscle segmentation and sarcopenia detection across heterogeneous abdominal CT protocols and diverse clinical conditions.

Given the limitations of conventional approaches in managing indeterminate thyroid nodules, there remains an unmet need for non-invasive assistant tools to improve risk stratification. This study aimed to evaluate the clinical applicability of an artificial intelligence (AI) model for thyroid nodules with atypia of undetermined significance (AUS) cytology. We retrospectively analyzed patients who underwent fine-needle aspiration (FNA) for thyroid nodules between January 2019 and December 2020 across five medical institutions in Korea. Nodules initially diagnosed as AUS and later confirmed as benign or malignant were included. A previously developed deep learning-based AI model, AI-Thyroid, was employed to provide binary classifications (benign or malignant) and malignancy risk estimates. A total of 165 thyroid nodules were analyzed. The median [interquartile range] longest diameter was 1.30 cm [0.80-2.10], and the malignancy rate of the cohort was 39%. In binary classification tasks, the model achieved a sensitivity of 0.91 and a negative predictive value of 0.87. The area under the curve (AUC) based on estimated malignancy risk was 0.75 (95% confidence interval: 0.68-0.83), and the AUC derived from K-TIRADS categories 2 to 5 was 0.76 (95% confidence interval: 0.69-0.83), indicating comparable diagnostic accuracy with the traditional scoring system. Subgroup analyses demonstrated that the model achieved a sensitivity of 98% in nodules smaller than 1.5 cm. AI-assisted ultrasound analysis offers supplementary diagnostic information for thyroid nodules with AUS cytology. Its high sensitivity and negative predictive value may assist clinicians in decision-making processes, particularly for small, low-risk thyroid nodules.

Aortic diseases, particularly acute aortic syndromes (AAS) and aortic aneurysms (AA), represent critical cardiovascular conditions with high mortality rates requiring precise imaging for diagnosis and management. This review provides a comprehensive analysis of current imaging diagnostic techniques, focusing specifically on acquired thoracic and abdominal aortic pathologies. We first evaluate the comparative efficacy of Computed Tomography Angiography (CTA) and Magnetic Resonance Imaging (MRI) in the diagnosis of AAS (including aortic dissection, intramural hematoma, and penetrating atherosclerotic ulcer), highlighting the role of artificial intelligence in optimizing segmentation and detection. Subsequently, we discuss aortic aneurysms, emphasizing the shift from simple diameter-based assessment to functional risk stratification incorporating calcification scoring, inflammatory imaging, and hemodynamic parameters. Furthermore, the review addresses postoperative imaging surveillance, particularly for endoleak detection following endovascular aneurysm repair (EVAR). We conclude that while CTA remains the gold standard for emergency diagnosis due to its speed and spatial resolution, MRI offers superior value in functional assessment and radiation-free long-term follow-up. The integration of multimodal imaging and AI-driven analysis is essential for achieving precision medicine in the management of acquired aortic diseases.

Repeat transurethral resection (ReTUR) is essential for reducing residual and recurrent non-muscle-invasive bladder cancer (NMIBC). Pathological upstaging after ReTUR significantly influences prognosis. This study aimed to develop an interpretable machine learning model using CT radiomics to predict the risk of pathological upstaging following ReTUR in NMIBC patients. We retrospectively analyzed 104 NMIBC patients who underwent ReTUR at the Second Affiliated Hospital of Nanchang University from March 2019 to July 2022. Data were split 7:3 into training and internal validation sets. An external validation set included 40 patients from two other hospitals. Radiomic features were extracted from preoperative CT scans. Least Absolute Shrinkage and Selection Operator (LASSO) and multivariate logistic regression were used to identify predictors of pathological upstaging. Four machine learning models, including Extreme Gradient Boosting (XGBoost), Gradient Boosting Decision Tree (GBDT), Random Forest (RF), and Linear Discriminant Analysis (LDA), were constructed and evaluated using AUC, accuracy, precision, F1 score, calibration curves, and decision curve analysis (DCA). The best model was interpreted via SHapley Additive exPlanations (SHAP) to identify key predictive features. umor grade (OR = 7.02, 95% CI: 1.17-42.21), tumor size (OR = 5.83, 95% CI: 1.21-28.15), and tumor number (OR = 6.83, 95% CI: 1.18-39.52) were independent risk factors. From 4,738 radiomic features, nine were selected. The XGBoost model outperformed others, with an AUC of 0.804 (95% CI: 0.756-0.862), accuracy of 77.4%, precision of 82.7%, and F1 score of 0.701 in internal validation. External validation confirmed its robustness. SHAP analysis highlighted Wavelet_LLH_firstorder_Maximum.1, Gradient_ngtdm_Complexity, and tumor grade as top predictors. The model showed good calibration and clinical utility on DCA. An interpretable CT radiomics-based machine learning model integrating clinical and imaging features was developed to accurately predict pathological upstaging risk after ReTUR in NMIBC patients. This tool may support clinical decision-making for individualized treatment after multicenter validation.