BACKGROUND AND OBJECTIVE:Real-time ultrasound bone visualization is crucial for orthopedic surgical navigation, but current methods rely on a two-stage "segmentation-then-rendering" pipeline that adds latency and separates the enhancement from the original image context. These methods also struggle with discontinuous bone contours due to acoustic shadowing and angle sensitivity, limiting their clinical utility. We aimed to develop a new end-to-end deep learning framework to overcome these challenges and provide real-time, continuous bone visualization. METHODS:We reformulated ultrasound bone visualization as a direct guided overlay task: a deep network predicts a calibrated bone probability map and fuses it with the ultrasound image, eliminating separate mask rendering. Based on this concept, we developed BoneContourNet-Vis, a lightweight end-to-end model built on a ConvNeXt V2 Nano backbone for strong feature extraction and high-throughput inference. The network incorporates three specialized modules: an Edge Attention Module (EAM) to enhance thin cortical edges, a Physics-aware Interaction Module (PIM) to inject acoustic shadow and phase priors into deep features, and a Contour-Adaptive Module (CAM) to enforce smooth, continuous bone contours via graph-based refinement. RESULTS:Comprehensive evaluation on the Bone100K dataset (∼100k ultrasound frames) demonstrated that our method outperforms representative approaches (e.g., HiFormer, MedNeXt, MedSAM) with Dice 0.933, IoU 0.875, Precision 0.96, Recall 0.91, Specificity 0.95, HD95 6.52 px, ASSD 2.07 px, while achieving ∼109 frames per second (9.18 ms latency) - meeting intraoperative real-time requirements. Our approach showed more complete and continuous bone contours under heavy acoustic shadowing and maintained the grayscale context of the original ultrasound. CONCLUSIONS:The proposed BoneContourNet-Vis framework improves the completeness and interpretability of ultrasound bone imaging without sacrificing inference speed. It delivers an accurate and real-time bone visualization solution. In future work, this framework can be extended to multi-planar or dynamic ultrasound sequences to achieve real-time three-dimensional bone reconstruction. Combined with optical tracking and probe calibration, the method can further enhance millimeter-level localization accuracy, thereby providing robust technical support for clinical orthopedic navigation, intraoperative guidance, and rapid bedside fracture assessment.

Polyp image segmentation is a critical task in medical image analysis, aiding doctors in accurate diagnosis and treatment by precisely identifying polyps within images. Recent advancements in deep learning, particularly convolutional neural networks (CNNs), have significantly advanced polyp segmentation. However, accurately delineating polyps remains challenging due to their indistinct boundaries and irregular shapes. In response to these challenges, we developed CSA-Net, a specialized polyp segmentation network featuring two Transformer-inspired modules: Neighborhood Information Aggregation (NIA) and Multi-Axis Frequency domain Aggregation (MAFA). These modules integrate a self-attention mechanism akin to Transformers into CNNs, enabling the model to capture both local and global features effectively. Notably, MAFA leverages spectral energy in the frequency domain to achieve faster self-attention and more precise control over global features. Addressing the issue of information loss from encoder to decoder in CNNs, we introduce two novel components: Hierarchical Feature Unit (HFU) and Reverse Boundary Enhancement (RBE). HFU facilitates multiscale interactions and guides segmentation across different scales through transformations and cascades. Meanwhile, RBE integrates local and global features to refine boundaries. Experimental results demonstrate the effectiveness of our approach in polyp segmentation. Specifically, on the Kvasir dataset, our model achieves promising performance metrics with mDice of 0.924 and mIoU of 0.868. These results underscore the potential of CSA-Net in advancing the field of polyp segmentation, offering enhanced accuracy and robustness in medical image analysis applications.

Objective To develop and validate a Union Model integrating clinical features with radiomics signatures derived from intratumoral and peritumoral T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) for the non-invasive prediction of p53 expression in intrahepatic mass-forming cholangiocarcinoma (IMCC).Furthermore, the study aims to determine the optimal predictive range. Methods This retrospective study included IMCC patients with immunohistochemically confirmed p53 expression status from four institutions. Patients were allocated into training, internal validation, and external validation cohorts. Multi-scale regions of interest, integrating the intratumoral volume with peritumoral expansions of 0, 3, 5, 10, 15, and 20 mm, were delineated on the MRI sequences. Radiomics and clinical features were extracted to construct single-sequence intratumoral and peritumoral models, a Fusion Model, a Clinical Model, and a Union Model. Performance was assessed using receiver operating characteristic (ROC) curves, area under the curve (AUC), accuracy, sensitivity, and specificity. Model interpretability was explored via SHapley Additive exPlanations (SHAP) values. Results The intratumoral and peritumoral 5mm models based on DWI and T2WI sequences demonstrated robust performance, yielding AUCs of 0.959 and 0.987 in the training cohort, and 0.821 and 0.809 in the internal validation cohort, respectively. The Union Model demonstrated the best predictive performance, with AUC values of 0.992, 0.945, and 0.825 across the three cohorts. Conclusion A multimodal Union Model incorporating intratumoral and 5-mm peritumoral MRI features can effectively predict p53 expression in IMCC, demonstrating favorable generalizability and potential for clinical decision-making.

BackgroundIn recent years, the incidence of mycoplasma pneumonia (MP) in children has gradually increased; however, to date, few studies have assessed its prognosis in children. Hence, the present study aimed to develop a prognostic model for children with MP by using clinical data and radiomics features extracted from chest computed tomography (CT) images.MethodsA total of 356 children with MP from two hospitals were enrolled in the study. These patients were randomly assigned to a training set (n = 206), a test set (n = 52), and a validation set (n = 98). Clinical data, including patients’ history and demographics, laboratory test results, and CT imaging features, were collected. Radiomics features of the infected lesions were extracted from chest CT images. Univariate analysis was performed to determine the most predictive features with significant differences (P < 0.05) among all imaging and clinical data. Least absolute shrinkage and selection operator was used to select radiomics features and estimate the prediction performance of the clinical factors model and the radiomics model for prognosis.ResultsIn the training set, the clinical factors (creatine kinase-MB, lactate dehydrogenase, C-reactive protein, bilateral lobe pneumonia, number of lobes, and lesion volume) showed a significant prognostic value (all P < 0.05) and were selected to construct the clinical model. The area under the curve (AUC) values of the clinical model for the training, validation, and test sets were 0.767, 0.731, and 0.685, respectively. Twenty one radiomics features were selected. The AUC values of the radiomics model for the training, validation, and test sets were 0.829, 0.775, and 0.701, respectively. Decision curve analysis showed that the radiomics model has potential clinical application value for predicting the prognosis of children with MP. The clinical and radiomics model showed the best prediction performance, with AUC values of 0.842, 0.825, and 0.748 for the training, validation, and test sets, respectively.ConclusionThe radiomics model based on CT imaging features showed potential as a quantitative tool to predict the prognosis of children with MP. The addition of radiomics features to the clinical factors improved the diagnostic efficiency of the clinical and radiomics model. Thus, the combination of clinical data and radiomic features effectively predicted the prognostic outcome of children with MP.

To develop an interpretable magnetic resonance imaging (MRI)-based framework for preoperative histologic grading of intrahepatic mass-forming cholangiocarcinoma (IMCC) and exploratory prognostic stratification. A retrospective analysis was conducted on preoperative MRI from 333 IMCC patients across three centers (training cohort, n = 240; external validation cohort, n = 93). An ensemble deep learning (DL) framework synergizing 2.5D and 3D ResNet-50 architectures was constructed. Significant variables from clinical-laboratory-imaging (ClinLabImag) features, radiomics, and DL outputs were integrated into a combined model. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC). Model interpretability was evaluated with Gradient-weighted Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP), and the Kaplan–Meier method was used to compare overall survival (OS) between risk groups. The DL model achieved an external validation AUC of 0.804 (95

ABSTRACT Objective Various machine learning and deep learning methods were used to develop and validate radiomics models based on magnetic resonance imaging (MRI) to predict eligibility for conservative management of endometrial cancer. Methods This retrospective study included 184 patients with early endometrial cancer confirmed by histopathology from two medical centers, who were divided into a training set (n = 123), an internal validation set (n = 31), and an external test set (n = 30). T2 weighted imaging (T2WI), dynamic contrast‐enhanced T1‐weighted imaging (DCE‐T1WI), and apparent diffusion coefficient (ADC) maps were used to extract radiomics features. A random forest model was used to select the most important features. Radiomics models were constructed using four machine learning methods including support vector machine (SVM), random forest, decision tree, logistic regression, and one deep learning method that is multilayer perceptron (MLP). They were validated using five‐fold cross validation and tested on the external test set. The performance of the models was evaluated by receiver operating characteristic curves, calibration curves, and clinical decision curves. Results The radiomics model established using MLP showed a certain performance in predicting eligibility for conservative management in early endometrial cancer patients. In the validation set, the area under the receiver operating characteristic curve (AUC) was 0.946 (95% CI 0.915–0.973). In the external test set, the area under the receiver operating characteristic curve (AUC) was 0.792 (95% CI 0.571–0.965). The calibration curve demonstrated good fit of the MLP model. The results of the DCA indicated that the MLP model had high net benefit in predicting eligibility for conservative management when the threshold probability ranged from 0.1 to 0.9. Conclusion The radiomics model based on MRI can be used to predict eligibility for conservative management in early endometrial cancer patients, providing objective imaging evidence for clinical decision‐making.

A 25-year-old male with tachycardia and right heart enlargement was diagnosed with atrial-ventricular nodal reentrant tachycardia (AVNRT) and successfully treated with radiofrequency ablation. Cardiac MRI confirmed right heart enlargement but ruled out cardiomyopathy. Further imaging revealed a rare congenital absence of the portal vein (CAPV) with an associated portosystemic shunt. This case highlights the importance of considering CAPV in unexplained cardiac symptoms and structural heart changes.

The performance of deep learning models for the early diagnosis of breast cancer based on mammograms often degrades when domain shifts occur. Although certain domain generalization methods can produce images with new styles for alleviating domain shifts, each produced image exhibits a single style. The present study introduces a Fourier transformation-based jigsaw puzzle (F-Jip) method which incorporates a jigsaw puzzle generation (JPG) module, a new style generation (NSG) module, and an enhanced jigsaw puzzle generation (EJPG) process to produce the enhanced jigsaw puzzles with multi-domain information. The enhanced jigsaw puzzles can guide the model to learn domain-invariant features to alleviate the influence of domain shifts for crossdomain breast cancer diagnosis. A leave-one-domain-out cross validation using six datasets which can simulate varying domain shifts encountered in clinical scenarios is used to evaluate the performance of the models. In each fold, the AUC and accuracy of the proposed model are higher than those of several state-of-the-art domain generalization models. Additionally, when compared to BI-RADS assessments of the radiologists in one dataset, the proposed model exhibits better performance for discriminating between the benign and malignant lesions, and further identifies 81.3 % patients with benign lesions in the subgroup analysis of patients with BI-RADS 4 lesions to avoid unnecessary biopsies. The experimental results indicate that the proposed model has the potentials for assisting doctors in diagnosing breast cancer.

BackgroundPatients with epidermal growth factor receptor (EGFR)-mutant lung adenocarcinoma who develop resistance to first-generation tyrosine kinase inhibitors (TKIs) without a T790M mutation face a therapeutic dilemma with limited and suboptimal options.MethodsIn this multicenter retrospective study, the final analyzable modeling cohort included 490 patients with complete eligible CT imaging and outcome labels, comprising a training cohort of 326 patients, validation cohort 1 of 70 patients, and validation cohort 2 of 94 patients. Model performance was evaluated using AUC, decision curve analysis, and PFS stratification analyses.ResultsThe 2.5D axial model achieved AUCs of 0.885, 0.819, and 0.863 in the training cohort, validation cohort 1, and validation cohort 2, respectively, based on the locked source prediction files.ConclusionA CT-based 2.5D deep learning model showed promising performance for treatment-response prediction after EGFR-TKI resistance, but prospective validation and clinical-variable benchmarking remain necessary before clinical implementation.

Rationale and Objectives This study aimed to develop and validate a deep learning-based brain metastasis detection model (BMDM) in magnetic resonance images for diagnosing brain metastases (BMs). Materials and Methods We retrospectively collected data from 950 patients serving as the training and test sets for developing BMDM and from an additional 423 patients as the validation set. Three reading modes were compared: radiologists only (10 total, four with ≤3 years of experience and six with >3 years of experience), BMDM only, and radiologists assisted by the BMDM. The alternative free-response receiver operating characteristic (AFROC) method was used for evaluation. Results The reading time was reduced by 30.87%, AFROC-area under the curve improved from 0.837 to 0.954, and sensitivity increased from 0.685 to 0.916 with BMDM assistance. The improvement in sensitivity was more pronounced among less experienced radiologists (24.59% vs 22.03%). The detection sensitivity improved by 33.45% for lesions ≤3 mm and by 43.00% for insular lesions. Conclusion The results demonstrated that BMDM significantly enhanced time efficiency and diagnostic performance for BM detection, providing clinical benefits.

Background: One of the most common primary tumor sources of brain metastases (BMs) is lung cancer. As certain magnetic resonance imaging (MRI) features overlap between the pathological and genetic subtypes of lung cancer BMs, directly determining the primary site based on these features remains a challenge. Thus, identifying the MRI features of different subtypes of lung cancer BMs is crucial in order to facilitate early diagnosis and treatment. This study aimed to characterize the MRI characteristics distinct to the various subtypes of lung cancer BMs in order to inform clinical decision-making. Methods: Data from 1,129 patients diagnosed with lung cancer BMs (a total of 8,312 lesions) from three institutions, including clinicopathological information and MRI features, were retrospectively analyzed. Among these cases of BMs, 369 (2,780 lesions) originated from small-cell lung cancer (SCLC) and 760 (5,532 lesions) from non-small cell lung cancer (NSCLC). Among the patients with NSCLC, there were 689 cases (5,243 lesions) of adenocarcinoma (AD) and 71 cases (289 lesions) of squamous cell carcinoma (SCC). Regarding epidermal growth factor receptor (EGFR) status, there were 188 wild-type cases (1,257 lesions) and 344 mutant-type cases (2,880 lesions). This study was divided into three parts. For Part I (comparison between SCLC and NSCLC), Part II (comparison between AD and SCC), and Part III (comparison between EGFR wild type and mutant type), a stepwise in-depth analysis was performed-from the level of pathological classification to the level of gene mutation status-of the clinical characteristics of patients with lung cancer BMs and of the quantity, size, location, and signal characteristics of BMs lesions based on brain MRI. According to different signal combinations of DWI and CE-T1WI, the BMs lesions were divided into seven patterns, namely Pattern I-VII: Pattern I: DWI-negative + CE-T1WI-positive; Pattern II: DWI-negative + CE-T1WI ring; Pattern III: DWI-positive + CE-T1WI-positive; Pattern IV: DWI ring + CE-T1WI-positive; Pattern V: DWI-positive + CE-T1WI ring; Pattern VI: DWI ring + CE-T1WI ring; Pattern VII: DWI-positive + CE-T1WI-negative. "Positive" indicates homogeneous hyperintensity on DWI or CE-T1WI, while "negative" indicates hypointensity or isointensity on DWI or CE-T1WI, and "ring" refers to ring enhancement. Results: In the Part I analysis, SCLC BMs tended to be multiple (>10 lesions; 0.5-1 cm in size), occur in the frontal/parietal lobes and periventricular regions, and have higher proportions of patterns consisting of diffusion-weighted imaging (DWI)-positive plus contrast-enhanced T1-weighted imaging (CE-T1WI) ring features, DWI ring plus CE-T1WI ring features, and DWI-positive plus C E-T1WI-negative features (all P values <0.05). NSCLC BMs had higher proportions of patterns consisting of DWI-negative plus CE-T1WI-positive features and DWI ring plus CE-T1WI-positive features (P<0.05). In the Part II analysis, as compared to AD BMs, SCC BMs have more peritumoral edema, and occur in the centrum semiovale, with higher proportions of patterns consisting of DWI ring plus CE-T1WI ring features and DWI-positive plus CE-T1WI-negative features (P<0.05). EGFR mutant-type BMs tended to be multiple (>10 lesions; <1 cm in size) and have less hemorrhage compared to wild-type BMs (P>0.05). DWI hyperintensity without CE-T1WI enhancement was more common in SCLC BMs than in NSCLC BMs (20.6% vs. 4.5%; P<0.001) and in SCC BMs than in AD BMs (33.9% vs. 2.8%; P<0.001). Conclusions: MRI and clinical features may provide the ability to noninvasively distinguish between SCLC and NSCLC and between AD and SCC, as well as to partially indicate EGFR status. DWI hyperintensity without CE-T1WI enhancement might serve as a key subtype-specific feature that could aid in clinical decision-making.

The aim of this study is to assess the predictive efficacy of a radiomics model utilizing magnetic resonance imaging (MRI) for predicting locally advanced cervical squamous cell carcinoma response to concurrent chemoradiotherapy. We enrolled a cohort of 139 patients diagnosed with stage IIB to IV cervical squamous cell carcinoma, based on the 2018 FIGO classification, who underwent concurrent chemoradiotherapy and pre-/post-treatment MRI examinations. These patients were divided into complete response and partial response groups. Prior to the initiation of treatment, the areas of interest within the lesion were delineated on T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and enhanced T1-weighted imaging (T1WI) sequences, from which radiomic features were extracted. The group was randomly divided into training (n = 111) and validation (n = 28) sets (8:2 ratio) to identify the optimal features. Logistic regression models were constructed to predict treatment response, with distinct models based on the following imaging modalities: enhanced T1WI, T2WI, DWI, a combination of T2WI and enhanced T1WI (joint model 1), and a combination of T2WI, DWI, and enhanced T1WI (joint model 2). Model fitness and predictive performance were assessed using receiver operating characteristic curves, while the clinical applicability of the models was analyzed using decision curve analysis. A cumulative count of 2,264 radiomic features was derived from the region of interest in each imaging sequence. Subsequently, 18, 16, 15, 16, and 13 optimal features were selectively identified from the five models. These selected features were employed to formulate a radiomics model designed for the prediction of treatment response. These selected features were used to construct individual radiomics models aimed at predicting treatment response. Subsequently, all models achieved AUCs > 0.8 in the validation set, with Joint Model 2 demonstrating the highest performance (AUC = 0.939, 95% CI: 0.826-1; sensitivity = 0.773, specificity = 0.833). No significant differences were observed between Joint Model 2 and other models (P > 0.05). The MRI-based radiomics model has high potential in effectively predicting the efficacy of concurrent chemoradiotherapy for locally advanced cervical squamous cell carcinoma.

To construct and validate a model based on clinical characteristics and magnetic resonance imaging (MRI) radiomics to predict 1-year efficacy of epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) in patients with EGFR-mutant non-small cell lung cancer (NSCLC) brain metastases (BMs). This study retrospectively analyzed data from 338 patients with EGFR-mutant NSCLC BMs from three centers, including MRI, clinical and pathological data, and radiological features. Based on the selected significant radiomic features from intratumoral regions extracted from CE-T1WI, while exploring the value of features in 3/5/8 mm peritumoral regions, seven commonly used machine learning algorithms were compared to select the optimal one for model construction, and the best algorithm was selected for model construction. In the model predicting 1-year therapeutic efficacy, clinical, radiomic, and combined models were constructed separately. The model performance was evaluated using receiver operating characteristic curves. The final development cohort comprised 285 patients from Center 1, while the external validation set included 57 patients from Centers 2 and 3. In the model predicting 1-year EGFR-TKIs efficacy, the random forest algorithm, which showed the best application, was used to construct the model. Compared with the radiomic and clinical models, the combined model exhibited superior area under the curve performance in the test set (0.756 vs. 0.644 vs. 0.668). In the external validation set, the combined model achieved an area under the curve of 0.743 (95

BackgroundThird-generation epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitor (TKI) resistance poses a significant therapeutic challenge in advanced lung adenocarcinoma. This study aimed to develop and validate a computed tomography (CT)-based habitat radiomics model for predicting response to immunochemotherapy in EGFR-mutant lung adenocarcinoma patients after TKI resistance.MethodsThis retrospective multicenter study enrolled 475 patients from two medical centers. Patients were allocated to train (N = 332) and external validation (N = 143) cohorts. Habitat imaging was performed using K-means clustering to partition tumors into three distinct subregions. Radiomic features were extracted from both whole-tumor volumes and habitat subregions. A combined model combining clinical, conventional radiomics, and habitat features was constructed using machine learning algorithms and validated through cross-validation and external testing. The primary endpoint was objective response rate (ORR) based on Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 criteria, and overall survival (OS) was used as a secondary endpoint.ResultsThe combined model demonstrated superior predictive performance with area under the curve (AUC) of 0.904 (95% CI: 0.871–0.937) in the train cohort and 0.890 (95% CI: 0.838–0.942) in the validation cohort, significantly outperforming the clinical model, conventional whole-tumor radiomics model, and habitat model (all P < 0.001). Moreover, Kaplan–Meier analysis based on the risk groups stratified by the combined model revealed significant survival differences, with high-risk groups showing markedly shorter overall survival in both cohorts (training HR = 3.688, validation HR = 2.823, both log-rank P < 0.0001).ConclusionThis study developed and externally validated a CT-based habitat radiomics model for predicting response to immunochemotherapy in EGFR-mutant lung adenocarcinoma after EGFR-TKI resistance. The combined model achieved improved predictive performance compared with single-modality approaches. These findings suggest that incorporating habitat-based features may enhance the characterization of intratumoral heterogeneity and improve treatment response prediction. Notably, the model demonstrated a high negative predictive value, suggesting its potential to reduce unnecessary treatment in predicted non-responders. Further prospective and multi-center validation is warranted.

Medical image multimodal registration is not only an indispensable processing step in medical image analysis but also plays a crucial role in disease diagnosis and treatment planning. However, the complex and unknown spatial deformation relationships between different organs and different modalities pose significant challenges to multimodal image registration. To address this problem, this study proposes an unsupervised and discriminator-free multimodal registration method based on a dual loss function—SA-HMT. Specifically, to address the challenge of cross-modal feature matching, the multi-scale skip Transformer module proposed in this study employs a hierarchical architecture to capture multi-scale deformation features. In the shallow network, the multi-scale skip pyramid module extracts modality-independent local structural features through parallel multi-branch convolution, effectively overcoming the differential expression of edges and textures across different modalities. In the deep network, the Transformer module establishes long-range dependencies via self-attention mechanism, enabling adaptive fusion of local deformation features with global semantics and effectively alleviating the matching difficulty of cross-modal structural features. In addition, this study further proposes a structure-aware deformable convolution module. The two-stage joint mechanism of "feature perception-offset generation" enhances the accuracy of feature matching through their progressive collaboration. The effectiveness of SA-HMT has been fully verified in five public data sets (covering chest and abdomen CT-MR, lung CT, brain CT-MR, cardiac MRI) and clinical abdominal data. Compared with the advanced method R2Net, our model achieves improvements in core indicators such as DSC, and the registration accuracy is generally comparable or better.

Objectives:This study aimed to develop a predictive model, based on radiomics, to assess the occurrence of extrapulmonary organ involvement and predict recovery durations in children with Mycoplasma pneumoniae pneumonia (MPP). Materials and methods:We retrospectively included 556 confirmed MPP patients from three medical centers between October 2022 and December 2024. Feature parameters were selected and weighted using Z-score normalization and LASSO. A logistic regression model was constructed to assess extrapulmonary organ involvement. Model performance was evaluated using the area under the curve (AUC), calibration curves, and decision curves, with comparisons between models conducted using the DeLong test. For predicting recovery duration, a separate model was developed based on selected features and was evaluated using mean squared error (MSE), the coefficient of determination (R2), and mean absolute error (MAE). Results:In the evaluation of the extrapulmonary organ involvement model, the Radiomics Model showed statistically significant differences when compared with both the Clinical Laboratory Model [(AUC = 0.73; 95% CI, 0.53-0.88) vs. (AUC = 0.67; 95% CI, 0.49-0.84), p < 0.05] and the Image Feature Model [(AUC = 0.73; 95% CI, 0.53-0.88) vs. (AUC = 0.65; 95% CI, 0.45-0.80), p < 0.05]. Significant differences were observed between the Clinical Laboratory Model and the Image Feature Model in the combined organ involvement group (p < 0.05), but no statistical difference was found in other groups (p > 0.05). The Integrated Model outperformed the Radiomics Model, Clinical Laboratory Model, and Image Feature Model, achieving the highest predictive performance (AUC = 0.94; 95% CI, 0.84-0.99), with all differences being statistically significant (p < 0.01). For predicting recovery duration of extrapulmonary organ involvement, the modified MSE was 6.0, the modified MAE was 1.9, and the modified R2 Score was 0.6825, indicating acceptable prediction performance. Conclusion:This study demonstrated that incorporating radiomics significantly improved the predictive accuracy of clinical laboratory parameters and imaging features for assessing extrapulmonary organ involvement and forecasting recovery durations in MPP patients. This approach provided an effective tool to enhance diagnostic efficiency for clinicians.

We employed a three-phase approach, culminating in a randomized controlled trial, to assess the efficacy of 3D-printed liver models in hepatobiliary surgical planning. Phase one involved developing and selecting 35 optimal 3DP models based on timeliness, cost, precision, and alignment with digital simulations. Phase two utilized deep learning algorithms to optimize the 3D reconstruction process, significantly enhancing efficiency and accuracy compared to manual segmentation. In phase three, a randomized controlled trial with 64 patients compared surgical outcomes between those planned with AI-enhanced physical 3DP models and those with traditional digital simulations. Results demonstrated that 3DP models were produced rapidly (3.52 h at $152 each) with high precision, AI-assisted reconstruction reduced processing time (303.5 vs. 557 min), and patients using AI-enhanced physical 3DP models experienced less intraoperative blood loss. Integrating deep learning with 3D printing offers a cost-effective, scalable method to enhance surgical planning and outcomes in hepatobiliary surgery.

PURPOSE:The present study aimed to develop a noninvasive predictive framework that integrates clinical data, conventional radiomics, habitat imaging, and deep learning for the preoperative stratification of MGMT gene promoter methylation in glioma. MATERIALS AND METHODS:This retrospective study included 410 patients from the University of California, San Francisco, USA, and 102 patients from our hospital. Seven models were constructed using preoperative contrast-enhanced T1-weighted MRI with gadobenate dimeglumine as the contrast agent. Habitat radiomics features were extracted from tumor subregions by k-means clustering, while deep learning features were acquired using a 3D convolutional neural network. Model performance was evaluated based on area under the curve (AUC) value, F1-score, and decision curve analysis. RESULTS:The combined model integrating clinical data, conventional radiomics, habitat imaging features, and deep learning achieved the highest performance (training AUC = 0.979 [95 % CI: 0.969-0.990], F1-score = 0.944; testing AUC = 0.777 [0.651-0.904], F1-score = 0.711). Among the single-modality models, habitat radiomics outperformed the other models (training AUC = 0.960 [0.954-0.983]; testing AUC = 0.724 [0.573-0.875]). CONCLUSION:The proposed multimodal framework considerably enhances preoperative prediction of MGMT gene promoter methylation, with habitat radiomics highlighting the critical role of tumor heterogeneity. This approach provides a scalable tool for personalized management of glioma.

ABSTRACT Objective To establish a model based on intratumoral and peritumoral radiomics for preoperatively differentiating solitary intrahepatic mass‐forming cholangiocarcinoma (IMCC) lesions from colorectal cancer liver metastases (CRLM). Methods Preoperative MRI scans from IMCC patients were retrospectively obtained from three academic medical centers. Radiomics features were extracted from the intratumoral and multiple peritumoral regions. After feature selection, the optimal peritumoral range was determined. The radiomics model was developed by integrating both single‐sequence and multisequence models through probabilistic ensemble learning. Significant variables from clinical imaging features, radiomics, were integrated into a combined model, and its performance was evaluated using the area under the receiver operating characteristic curve (AUC). Results A total of 170 patients (93 IMCC, 77 CRLM) comprised the training cohort, and 42 (23 IMCC, 19 CRLM) formed the external validation cohort. The Combined model achieved superior AUCs in training (0.978 [95% CI: 0.971–0.985]) and validation (0.940 [0.899–0.968]), outperforming radiomics (training: 0.947 [(95% CI: 0.932–0.961]); validation: 0.908 [(95% CI: 0.861–0.943)]) and Clin‐Imag models (training: 0.858 [95% CI: 0.831–0.880]; validation: 0.842 [95% CI: 0.786–0.889]) (DeLong test, p = 0.001). SHAP analysis identified DWI‐based 5‐mm peritumoral features and clinical‐imaging variables (e.g., lesion location and bile duct dilation) as key discriminators. Conclusions The combined model integrating clinical‐imaging variables and multiparametric MRI‐derived intratumoral and 5‐mm peritumoral radiomics features provides a non‐invasive tool for distinguishing solitary IMCC from CRLM, offering potential clinical utility for guiding personalized treatment strategies and avoiding unnecessary invasive interventions.