To develop and compare multiregional 2D and 2.5D deep learning models based on DCE-MRI for noninvasive prediction of axillary lymph node (ALN) pathological complete response (pCR) after neoadjuvant chemotherapy (NAC). This retrospective study enrolled 305 patients with invasive breast cancer and ipsilateral ALN metastasis, which were randomly assigned to a training set (n = 214) and a validation set (n = 91). The Mann–Whitney U test, Spearman correlation analysis, max-relevance and min-redundancy and least absolute shrinkage and selection operator were used to select the most significant features. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve analysis, and decision curve analysis. Among the 305 patients, ALN pCR accounted for 46.6

Objectives : To develop and compare multi-regional 2D and 2.5D deep learning models based on DCE-MR for noninvasive prediction of axillary pathological complete response (pCR) after neoadjuvant chemotherapy (NAC). Methods : This retrospective study enrolled 305 patients with invasive breast cancer and ipsilateral ALN metastasis, which were randomly assigned to a training set (n = 214) and a validation set (n = 91). The principal component analysis, U test, Spearman correlation analysis, mRMR and least absolute shrinkage and selection operator were used to select the most significant features. Seven supervised predictive models were constructed based on regions of interest from the primary tumor and ALNs. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve analysis, and decision curve analysis. Results : Among the 305 patients, axillary pCR accounted for 46.6% (142/305). ER status, tumor MP grade and HER2 status were selected as independent predictors for axillary pCR(P < 0.05). Both the 2.5D T and the 2.5D T + ALN models had higher AUC than those of the 2.5D T and 2D T + ALN models in the validation set (AUC: 0.797 vs 0.706, 0.834 vs 0.815). When combining clinicopathological factors, the 2.5D T + ALN +Clinic model achieved the highest performance among all models (AUC = 0.861). Model-guided decision-making reduced the estimated rate of unnecessary ALND from 40.7% to 8.8% and increased the overall clinical benefit rate from 59.3% to 74.7%. Conclusion The 2.5D T + ALN +Clinic model could accurate, noninvasive prediction of axillary pCR after NAC, which has the potential to support individualized axillary surgical decision-making.

PurposeTo develop and validate a preoperative MRI-based habitat radiomics nomogram for noninvasive prediction of axillary pathological complete response (apCR) after neoadjuvant therapy (NAT) in node-positive breast cancer.Patients and methodsThis retrospective multicenter study included patients with histologically confirmed node-positive breast cancer from two institutions who underwent pretreatment breast MRI. Dynamic contrast-enhanced MRI was used for tumor segmentation and to define intratumoral habitat subregions based on enhancement heterogeneity. Radiomic features were extracted from whole tumors and habitat subregions, and radiomics and habitat signatures were integrated with clinicopathologic variables to construct a nomogram for preoperative prediction of apCR. Model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration assessment, and decision curve analysis (DCA).ResultsA total of 336 women were included. In the training cohort, the radiomics, habitat, and nomogram models achieved AUCs of 0.723, 0.765, and 0.845, respectively. The nomogram consistently demonstrated the highest discriminative performance in the internal validation and independent external test cohorts, with AUCs of 0.755 and 0.754. Calibration analysis showed good agreement between predicted and observed apCR, and DCA indicated greater net clinical benefit for the nomogram.ConclusionThe proposed MRI-based habitat radiomics nomogram showed promising performance for noninvasive, preoperative prediction of apCR after NAT and may assist individualized axillary risk stratification in patients with node-positive breast cancer.

Purpose: To develop a deep learning radiomics (DLR) model based on longitudinal multiparametric breast MRI to predict axillary lymph node (ALN) response following neoadjuvant therapy (NAT) in breast cancer patients. Patients and Methods: This single-center retrospective study included 254 breast cancer patients who underwent NAT followed by surgery from January 2017 to October 2023. Pre-and post-NAT multiparametric MRI scans were analyzed to extract radiomics and deep learning features. The dataset was randomly divided into a training cohort (n = 144) and a validation cohort (n = 110). Feature selection was performed using the Mann-Whitney U-test, Spearman correlation analysis, and least absolute shrinkage and selection operator regression. Eight machine learning algorithms were compared, with logistic regression selected as the final classifier. Four models were constructed: clinical, radiomics, deep learning, and the DLR model. Performance was evaluated using ROC analysis, calibration curves, and decision curve analysis. Results: Estrogen receptor status, HER2 status, and clinical T stage were independent predictors of axillary pathological complete response (apCR). The DLR model achieved the highest predictive performance, with AUCs of 0.939 (95% CI: 0.905-0.974) in the training set and 0.856 (95% CI: 0.774-0.938) in the validation set. DeLong tests showed that the DLR model outperformed only the clinical model (p < 0.0001). A bootstrap analysis (2000 iterations) further showed that the AUC difference between the training and validation cohorts was statistically significant (difference = 0.083; 95% CI: 0.0019-0.1786; p = 0.043). Conclusion: This study is among the first to integrate longitudinal multiparametric MRI with deep learning-based radiomics for predicting ALN response after NAT. The proposed DLR model may provide a noninvasive aid to individualized axillary decision-making, pending external validation.

Background:Multimodal large language models are increasingly used in radiological diagnosis, but their performance has not been systematically evaluated across volumetric (3D) imaging, real-world clinical versus public teaching cases, and bilingual contexts. Objective:The aim of the study is to develop a bilingual radiology benchmark and characterize the diagnostic performance of state-of-the-art multimodal large language models across input modality, clinical setting (public teaching vs routine clinical), disease rarity, and clinical-history language and to disentangle linguistic from clinical-content effects through a cross-linguistic control experiment. Methods:We constructed RadM-Bench, comprising 720 cases evenly distributed across 9 radiological subspecialties: 360 English public teaching cases enriched in rare diseases (RadEdu) and 360 Chinese routine clinical cases (RealClin). In total, 4 proprietary models (GPT-4o, O3, Gemini-2-Flash, and Gemini-2.5-Flash-Thinking) and 6 open-source models (Qwen2.5-VL-72B/7B, InternVL3-78B/8B, Llama-4-Scout-17B-16E, and MedGemma-4B) were evaluated under 4 input conditions: clinical history alone, history with radiologist-selected 2D key images, and history with volumetric data sampled at 2 and 10 frames per second (fps). Each response was scored on a 4-tier 0-3 diagnostic-quality rubric by 2 board-certified radiologists blinded to model identity. Mean scores with bias-corrected and accelerated bootstrap 95% CIs are reported. To disentangle language from clinical content, all 360 RealClin histories were translated into English and re-evaluated, with paired comparisons by Wilcoxon signed-rank tests and Benjamini-Hochberg false-discovery-rate correction. Results:Mean performance remained below 1.5 on the 0-3 scale for all 10 models on both datasets. Adding radiologist-selected 2D key images to clinical history improved performance in all 10 models (+19.8% to +139.2%). In RealClin at fps=10, all 8 evaluable models scored lower with volumetric input than with the 2D-image baseline (-5.3% to -31.4%); MedGemma-4B and Llama-4-Scout-17B-16E could only be evaluated at fps=2 due to context-window and graphics processing unit-memory constraints. At fps=2, a total of 8 out of 10 models declined (-6.8% to -28.6%), while Qwen2.5-VL-7B and InternVL3-8B showed marginal improvements (+2.4% and +2.1%). Cross-dataset transfer diverged by model category: proprietary models declined from RadEdu to RealClin (eg, O3 with images: 1.14 to 0.79), whereas Chinese-centric open-source models improved (eg, InternVL3-78B: 0.48 to 0.75). The rare-disease premium observed in 9 out of 10 models in RadEdu reversed in RealClin, where common-disease scores exceeded rare-disease scores in 7 out of 10 models under history-only input. Translating RealClin histories into English produced a numerical decrease in mean score for all 10 models, which were statistically significant in 9 out of 10 models after false discovery rate correction, excluding a Chinese-language penalty. Conclusions:Within the scope of this benchmark, multimodal inputs improved performance over clinical history alone, but performance gaps remain in volumetric data processing and cross-context generalization, with mean diagnostic performance across the 10 evaluated models remaining below clinically actionable levels on both datasets.

We developed an artificial intelligence system (AIS) using multi-view multi-level convolutional neural networks for breast cancer detection, diagnosis, and BI-RADS categorization support in mammography. Twenty-four thousand eight hundred sixty-six breasts from 12,433 Asian women between August 2012 and December 2018 were enrolled. The study consisted of three parts: (1) evaluation of AIS performance in malignancy diagnosis; (2) stratified analysis of BI-RADS 3–4 subgroups with AIS; and (3) reassessment of BI-RADS 0 breasts with AIS assistance. We further evaluate AIS by conducting a counterbalance-designed AI-assisted study, where ten radiologists read 1302 cases with/without AIS assistance. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, and F1 score were measured. The AIS yielded AUC values of 0.995, 0.933, and 0.947 for malignancy diagnosis in the validation set, testing set 1, and testing set 2, respectively. Within BI-RADS 3–4 subgroups with pathological results, AIS downgraded 83.1

Background:Histological grade is an acknowledged prognostic factor for breast cancer, essential for determining clinical treatment strategies and prognosis assessment. Our study aims to establish intra- and peritumoral radiomics models using T2WI and DWI MR sequences for predicting the histological grade of breast cancer. Methods:700 breast cancer cases who had MRI scans before surgery were included. The intratumoral region (ITR) of interest was manually delineated, while the peritumoral region (PTR-3 mm) was automatically obtained by expanding the ITR by 3 mm. Radiomics features were extracted using the intra- and peritumoral images from T2WI and DWI sequences on breast MRI. Then, the key features with the strongest predictivity of histological grade were selected. Finally, 9 predictive radiomics models were established based on T2WI-ITR, T2WI-3mmPTR, DWI-ITR, DWI-3mmPTR, T2WI-ITR + 3mmPTR, DWI-ITR + 3mmPTR, (T2WI + DWI)-ITR, (T2WI + DWI)-3mmPTR and (T2WI + DWI)-ITR + 3mmPTR. Results:The (T2WI + DWI)-ITR + 3mmPTR contained 13 DWI features which included a shape feature, a texture feature, and 11 filtered features, as well as 10 T2WI features, all of which were filtered features. Among the 9 models, the combined models showed better performance than the single models in both the training and test sets, especially for the (T2WI + DWI)-ITR + 3mmPTR radiomics model. The (T2WI + DWI)-ITR + 3mmPTR radiomics model achieved a sensitivity, specificity, accuracy, and AUC of 80.4%, 72.4%, 75.0%, and 0.860 in the training set, and 68.9%, 70.5%, 70.0%, and 0.781 in the test set. Decision curve analysis (DCA) showed that the (T2WI + DWI)-ITR + 3mmPTR model had the greatest net clinical benefit compared to the other models. Conclusion:The intra- and peritumoral radiomics methodologies using T2WI and DWI MR sequences could be utilized to assess histological grade for breast cancer, particularly with the (T2WI + DWI)-ITR + 3mmPTR radiomics model demonstrating significant potential for clinical application.

Purpose: To investigate the potential of radiomics signatures (RSs) from intratumoral and peritumoral regions on multiparametric magnetic resonance imaging (MRI) to noninvasively evaluate HER2 status in breast cancer. Method: In this retrospective study, 992 patients with pathologically confirmed breast cancers who underwent preoperative MRI were enrolled. The breast cancer lesions were segmented manually, and the intratumor region of interest (ROIIntra) was dilated by 2, 4, 6 and 8 mm (ROIPeri2mm, ROIPeri4mm, ROIPeri6mm, and ROIPeri8mm, respectively). Quantitative radiomics features were extracted from dynamic contrast-enhanced T1-weighted imaging (DCE-T1), fat‐saturated T2‐weighted imaging (T2) and diffusion-weighted imaging (DWI). A three-step procedure was performed for feature selection, and RSs were constructed using a support vector machine (SVM) to predict HER2 status. Result: The best single-area RSs for predicting HER2 status were DCE_Peri4mm-RS, T2_Peri4mm-RS, and DWI_Peri4mm-RS, yielding areas under the curve (AUCs) of 0.716 (95% confidence interval (CI), 0.648–0.778), 0.706 (95% CI, 0.637–0.768), and 0.719 (95% CI, 0.651–0.780), respectively, in the test set. The optimal RSs combining intratumoral and peritumoral regions for evaluating HER2 status were DCE-T1_Intra + DCE_Peri4mm-RS, T2_Intra + T2_Peri6mm-RS and DWI_Intra + DWI_Peri4mm-RS, with AUCs of 0.752 (95% CI, 0.686–0.810), 0.754 (95% CI, 0.688–0.812) and 0.725 (95% CI, 0.657–0.786), respectively, in the test set. Combining three sequences in the ROIIntra, ROIPeri2mm, ROIPeri4mm, ROIPeri6mm and ROIPeri8mm areas, the optimal RS was DCE-T1_Peri4mm + T2_Peri4mm + DWI_Peri4mm-RS, achieving an AUC of 0.795 (95% CI, 0.733–0.849) in the test set. Conclusion: This study systematically explored the influence of the intratumoral region, different peritumoral sizes and their combination in radiomics analysis for predicting HER2 status in breast cancer based on multiparametric MRI and found the optimal RS.

ObjectiveTo investigate the value of predicting axillary lymph node (ALN) metastasis based on intratumoral and peritumoral dynamic contrast-enhanced MRI (DCE-MRI) radiomics and clinico-radiological characteristics in breast cancer.MethodsA total of 473 breast cancer patients who underwent preoperative DCE-MRI from Jan 2017 to Dec 2020 were enrolled. These patients were randomly divided into training (n=378) and testing sets (n=95) at 8:2 ratio. Intratumoral regions (ITRs) of interest were manually delineated, and peritumoral regions of 3 mm (3 mmPTRs) were automatically obtained by morphologically dilating the ITR. Radiomics features were extracted, and ALN metastasis-related radiomics features were selected by the Mann-Whitney U test, Z score normalization, variance thresholding, K-best algorithm and least absolute shrinkage and selection operator (LASSO) algorithm. Clinico-radiological risk factors were selected by logistic regression and were also used to construct predictive models combined with radiomics features. Then, 5 models were constructed, including ITR, 3 mmPTR, ITR+3 mmPTR, clinico-radiological and combined (ITR+3 mmPTR+ clinico-radiological) models. The performance of models was assessed by sensitivity, specificity, accuracy, F1 score and area under the curve (AUC) of receiver operating characteristic (ROC), calibration curves and decision curve analysis (DCA).ResultsA total of 2264 radiomics features were extracted from each region of interest (ROI), 3 and 10 radiomics features were selected for the ITR and 3 mmPTR, respectively. 5 clinico-radiological risk factors were selected, including lesion size, human epidermal growth factor receptor 2 (HER2) expression, vascular cancer thrombus status, MR-reported ALN status, and time-signal intensity curve (TIC) type. In the testing set, the combined model showed the highest AUC (0.839), specificity (74.2%), accuracy (75.8%) and F1 Score (69.3%) among the 5 models. DCA showed that it had the greatest net clinical benefit compared to the other models.ConclusionThe intra- and peritumoral radiomics models based on DCE-MRI could be used to predict ALN metastasis in breast cancer, especially for the combined model with clinico-radiological characteristics showing promising clinical application value.

ObjectiveTo establish and validate a new clinical-radiomics nomogram based on the fat-suppressed T2 sequence for differentiating luminal and non-luminal breast cancer.MethodsA total of 593 breast cancer patients who underwent preoperative breast MRI from Jan 2017 to Dec 2020 were enrolled, which were randomly divided into the training (n=474) and test sets (n=119) at the ratio of 8:2. Intratumoral region (ITR) of interest were manually delineated, and peritumoral regions of 3 mm and 5 mm (PTR-3 mm and PTR-5 mm) were automatically obtained by dilating the ITR. Intratumoral and peritumoral radiomics features were extracted from the fat-suppressed T2-weighted images, including first-order statistical features, shape features, texture features, and filtered features. The Mann-Whitney U Test, Z score normalization, K-best method, and least absolute shrinkage and selection operator (LASSO) algorithm were applied to select key features to construct radscores based on ITR, PTR-3 mm, PTR-5 mm, ITR+PTR-3 mm and ITR+ PTR-5 mm. Risk factors were selected by univariate and multivariate logistic regressions and were used to construct a clinical model and a clinical-radiomics model that presented as a nomogram. The performance of models was assessed by sensitivity, specificity, accuracy, the area under the curve (AUC) of receiver operating characteristic (ROC), calibration curves, and decision curve analysis (DCA).ResultsITR+PTR-3 mm radsore and histological grade were selected as risk factors. A clinical-radiomics model was constructed by adding ITR+PTR-3mm radscore to the clinical factor, which was presented as a nomogram. The clinical-radiomics nomogram showed the highest AUC (0.873), sensitivity (72.3%), specificity (78.9%) and accuracy (77.0%) in the training set and the highest AUC (0.851), sensitivity (71.4%), specificity (79.8%) and accuracy (77.3%) in the test set. DCA showed that the clinical-radiomics nomogram had the greatest net clinical benefit compared to the other models.ConclusionThe clinical-radiomics nomogram showed promising clinical application value in differentiating luminal and non-luminal breast cancer.

Background: Core biopsy sampling may not fully capture tumor heterogeneity. Radiomics provides a non-invasive method to assess tumor characteristics, including both the core and surrounding tissue, with the potential to improve the accuracy of HER-2 status prediction. Objective: To explore the clinical value of intratumoral and peritumoral radiomics features from dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) for preoperative prediction of human epidermal growth factor receptor-2 (HER-2) expression status in breast cancer. Methods: Two tasks were designed, including Task1-distinguished HER-2 positive and HER-2 negative from 382 breast cancer patients and Task2-distinguished HER-2 low and HER-2 zero expression from 249 patients with HER-2 negative. Three radiomics models (intratumoral, peritumoral 5 mm, intratumoral+peritumoral 5 mm) were constructed based on decision tree, and clinical combined radiomics models were constructed with logistic regression based on clinicopathological features and radscore. The area under the curve (AUC), sensitivity, specificity, accuracy and decision curve analysis (DCA) were used to evaluate the predictive performance of models. Results: Estrogen receptor (ER), progesterone receptor (PR) and Ki67 showed statistically significant in the different groups of HER- 2 expression. Additionally, magnetic resonance imaging-reported axillary lymph nodes (MRI-reported ALN) in the positive and negative groups and histological grade in the low and zero expression groups showed significant differences (all P < 0.05). For task 1, the peritumoral radiomics model outperformed the other two radiomics models, with AUC values of 0.774 and 0.727 in the training and testing sets, respectively. For task 2, intratumoral + peritumoral radiomics model in the testing set showed the best predictive performance among the three radiomics models, and the AUC values were 0.777. The addition of clinicopathological features slightly altered the AUC values in both tasks. Conclusion: Both radiomics methods based on DCE-MRI and the nomogram are helpful for preoperative prediction of HER-2 expression status.

Mammography is a widely used screening tool for breast cancer, and accurate diagnosis is critical for the effective management of breast cancer. In this study, we propose a novel cross-view mutual learning method that leverages a Cross-view Masked Autoencoder (CMAE) and a Dual-View Affinity Matrix (DAM) to extract cross-view features and facilitate malignancy classification in mammography. CMAE aims to extract the underlying features from multi-view mammography data without relying on lesion labeling information or multi-view registration. DAM helps overcome the limitations of single-view models and identifies unique patterns and features in each view, thereby improving the accuracy and robustness of breast tissue representations. We evaluate our approach on a large-scale in-house mammography dataset and demonstrate promising results compared to existing methods. Additionally, we perform an ablation analysis to investigate the influence of different loss functions on the performance of our method. The results show that all the proposed components contribute positively to the final performance. In summary, the proposed cross-view mutual learning method shows great potential for assisting malignant classification.

Objective:To investigate the value of intratumoral and peritumoral radiomics features of multi-parameter MRI in evaluation of the status of human epithelial growth factor receptor 2 in breast cancer.Methods:The clinical, pathological and imaging data of 340 patients with pathologically confirmed breast cancer in Henan Provincial People′s Hospital from September 2019 to December 2020 were retrospectively collected. All patients were female, 48 (42, 55) years old. All patients underwent multi-parameter breast MRI before surgery, including dynamic contrast-enhanced T 1WI (DCE-T 1WI), fat-suppressed T 2WI (T 2WI) and diffusion-weighted imaging (DWI). The region of interest (ROI) for lesions were manually delineated and the segmented ROIs were zoomed in ring shape by 4 mm to acquire ROI intra and ROI prei, respectively. Then six sets of radiomics features were extracted from ROI intra and ROI prei of DCE-T 1WI, T 2WI and DWI. The cases were divided into a training set (272 cases) and a test set (68 cases) by stratified sampling at a ratio of 4∶1. The Mann-Whitney U test, Select K Best and minimum absolute contraction and selection operator were used for feature selection of the 6 sets of radiomics features. The feature subsets after reduction were used to construct independent and combined radiomics signatures with support vector machine algorithm to predict the HER2 status of breast cancer. Receiver operating characteristic curve was generated and area under curve (AUC) was calculated to compare the prediction performance of different models. Results:Of the 340 patients, 80 were HER2-positive and 260 were HER2-negative. Among the radiomics signatures based on single sequence, the DWI peri showed the best performance in predicting HER2 status of breast cancer, with an AUC of 0.678 for the test set. Among the combination of intratumoral and peritumoral radiomics signatures based on same sequence, the DWI intra+DWI peri had the highest prediction value, achieving an AUC of 0.774 for the testing set. Among the intratumoral or peritumoral radiomics signatures derived from two different sequences, the DCE-T 1WI intra+DWI intra and T 2WI peri+DWI peri showed the best predictive performance, yielding AUC of 0.766 and 0.769 in the testing set, respectively. Among the combination of intratumoral or peritumoral radiomics signatures derived from all 3 sequences or combinations of all features, the DCE-T 1WI intra+T 2WI intra+DWI intra+DCE-T 1WI peri+T 2WI peri+DWI peri obtained the highest prediction efficiency, with an AUC of 0.913 for the testing set. Conclusion:The radiomics features of intratumoral and peritumoral regions based on multi-parameter MRI have a certain value in non-invasive evaluation of HER2 status of breast cancer, which can help clinicians to provide scientific basis for decision-making of targeted therapy in patients with breast cancer.

目的 探讨临床病理及病灶MRI特征与乳腺癌腋窝淋巴结(ALN)转移的相关性及其辅助提高ALN转移诊断的价值.方法 回顾性分析842 例病理证实为乳腺癌患者的临床病理及MRI特征,根据病理结果将患者分为ALN转移组和无转移组;分析两组患者在临床病理及MRI特征方面的差异,并用Logistic回归分析乳腺癌ALN转移的独立影响因素,绘制受试者工作特征曲线(ROC)评价其诊断效能.结果 842 例患者中ALN转移组 307例,无转移组535 例;病理分级、孕激素受体(PR)、人表皮生长因子受体-2(HER-2)、Ki-67 及是否合并脉管癌栓在两组患者中差异均有统计学意义(P均<0.05).MRI特征中病灶大小、数目、强化类型及乳腺影像报告和数据系统(BI-RADS)分类在两组患者中差异均有统计学意义(P均<0.05);以MRI上ALN形态学标准诊断ALN转移,本研究共检出可疑ALN转移患者 390 例(46.3%),无转移者 452 例(53.7%),其灵敏度、特异度、准确率分别为82.1%、74.2%和77.1%.Logistic回归分析显示PR低表达、HER-2 阳性、合并脉管癌栓、病灶>20 mm及非肿块强化、BI-RADS 5 类是乳腺癌ALN转移的独立危险因素(P均<0.05);基于MRI形态学标准、临床病理及病灶MRI特征、及二者联合诊断乳腺癌ALN转移的曲线下面积(AUC)值分别为0.781、0.739 和0.848.结论 临床病理及病灶MRI特征与乳腺癌ALN转移有一定的相关性,其与MRI上ALN形态学诊断标准联合可提高乳腺癌ALN转移的诊断效能.

To develop an [18F]FDG PET/3D-UTE model based on clinical factors, three-dimensional ultrashort echo time (3D-UTE), and PET radiomics features via machine learning for the assessment of lymph node (LN) status in non-small cell lung cancer (NSCLC). A total of 145 NSCLC patients (training, 101 cases; test, 44 cases) underwent whole-body [18F]FDG PET/CT and chest [18F]FDG PET/MRI were enrolled. Preoperative clinical factors and 3D-UTE, CT, and PET radiomics features were analyzed. The Mann–Whitney U test, LASSO regression, and SelectKBest were used for feature extraction. Five machine learning algorithms were used to establish prediction models, which were evaluated by the area under receiver-operator characteristic (ROC), DeLong test, calibration curves, and decision curve analysis (DCA). A prediction model based on random forest, consisting of four clinical factors, six 3D-UTE, and six PET radiomics features, was used as the final model for PET/3D-UTE. The AUCs of this model were 0.912 and 0.791 in the training and test sets, respectively, which not only showed different degrees of improvement over individual models such as clinical, 3D-UTE, and PET (AUC-training = 0.838, 0.834, and 0.828, AUC-test = 0.756, 0.745, and 0.768, respectively) but also achieved the similar diagnostic efficacy as the optimal PET/CT model (AUC-training = 0.890, AUC-test = 0.793). The calibration curves and DCA indicated good consistency (C-index, 0.912) and clinical utility of this model, respectively. The [18F]FDG PET/3D-UTE model based on clinical factors, 3D-UTE, and PET radiomics features using machine learning methods could noninvasively assess the LN status of NSCLC. A machine learning model of 18F-fluorodeoxyglucose positron emission tomography/ three-dimensional ultrashort echo time could noninvasively assess the lymph node status of non-small cell lung cancer, which provides a novel method with less radiation burden for clinical practice. • The 3D-UTE radiomics model using the PLS-DA classifier was significantly associated with LN status in NSCLC and has similar diagnostic performance as the clinical, CT, and PET models. • The [ 18 F]FDG PET/3D-UTE model based on clinical factors, 3D-UTE, and PET radiomics features using the RF classifier could noninvasively assess the LN status of NSCLC and showed improved diagnostic performance compared to the clinical, 3D-UTE, and PET models. • In the assessment of LN status in NSCLC, the [ 18 F]FDG PET/3D-UTE model has similar diagnostic efficacy as the [ 18 F]FDG PET/CT model that incorporates clinical factors and CT and PET radiomics features.

目的 探讨外周血淋巴细胞亚群及MRI特征与乳腺癌Luminal分型的相关性.方法 回顾性分析 253例Luminal型乳腺癌外周血淋巴细胞亚群及其MRI特征,以ER阳性/HER-2 阴性/Ki-67<20%/PR>20%作为Luminal A型乳腺癌分型标准将患者分成Luminal A型和B型两组.分析临床病理、外周血淋巴细胞亚群及乳腺MRI等特征在两组患者中的差异,用Logistic回归分析乳腺癌Luminal B型的独立影响因素及其预测效能.结果 (1)253 例Luminal型乳腺癌患者中Luminal A型61 例,Luminal B型192 例;病理SBR分级高、腋窝淋巴结转移及脉管侵犯的患者更易出现在Luminal B型中,且差异均有统计学意义(P值均<0.05).(2)Luminal B型乳腺癌辅助T淋巴细胞绝对数的中位数较Luminal A型乳腺癌的低(691 vs 773),且差异有统计学意义(P =0.033).(3)MRI上病灶越大、BI-RADS分类越高及腋窝淋巴结转移的患者易出现在Luminal B型乳腺癌患者中,且差异均有统计学意义(P值均<0.05).(4)Logistic回归分析结果表明病理SBR 3 级及MRI上病灶≥20 mm是Luminal B型乳腺癌的独立影响因素,其OR值分别为22.182 和1.981;基于病理分级及MRI上病灶大小及二者联合预测Luminal B型乳腺癌的曲线下面积分别为0.678、0.623 和0.740.结论 乳腺癌的临床病理、外周血淋巴细胞亚群及MRI特征与Luminal分型有一定的相关性,且病理分级及MRI上病灶大小是Luminal B型乳腺癌的独立影响因素.

目的 探讨动态对比增强MRI(dynamic contrast enhanced MRI,DCE-MRI)影像组学特征在预测乳腺癌腋窝淋巴结(axillary lymph node,ALN)转移中的价值.材料与方法 回顾性分析2017年1月至2020年12月间经河南省人民医院手术病理证实为乳腺癌患者的首次术前MRI图像及临床病理资料(包括患者年龄、病灶的位置和大小、SBR分级,ER、PR、HER-2及Ki-67的表达情况,ALN是否转移及脉管癌栓的有无),共入组356例患者,年龄26～82(49.17±10.75)岁,并按8∶2的比例将其随机分为训练集(n=284)和测试集(n=72).提取DCE-T1WI序列第3期图像的影像组学特征,并通过Mann-Whitney U检验、Z分数归一化、方差阈值、K最佳及最小绝对值收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法筛选与ALN转移强相关的定量影像组学特征;应用多种分类器算法以排列组合方式分别构建影像组学标签,并利用受试者工作特征(receiver operating characteristic,ROC)曲线得到的曲线下面积(area under the curve,AUC)、敏感度、特异度及准确度评价模型的预测性能,根据模型效能从中确定最佳影像组学预测模型.结果 356例乳腺癌患者中ALN转移组117例(32.9％,117/356),无ALN转移组239例(67.1％,239/356);HER-2阳性表达在ALN转移组和无转移组之间的差异有统计学意义(χ2=5.433,P=0.020),其余临床病理指标在两组间的差异均无统计学意义(P＞0.05);且临床病理指标在训练集与测试集患者中的差异均无统计学意义(P＞0.05).从初始653个影像组学特征中共筛选得到18个与ALN转移强相关的影像组学特征,包括形态特征、一阶特征及纹理特征各6个.基于绝对值最大归一化和Bagging决策树算法构建的影像组学标签是预测ALN转移的最佳模型,该模型在训练集和测试集的AUC、敏感度、特异度和准确度分别为0.929[95％置信区间(confidence interval,CI):0.897～0.960]、69.9％、96.9％、88.0％和0.803(95％CI:0.701～0.905)、75.0％、75.0％、75.0％.结论 基于DCE-MRI影像组学特征构建的预测模型有助于乳腺癌术前ALN评估.

在全球范围内,乳腺癌发病率居于前列,严重威胁妇女身心健康,早期诊断可显著提高乳腺癌患者的生存率.近年来随着大数据及计算机算法的发展,影像组学和深度学习等人工智能(artificial intelligence,AI)技术在医学影像领域中的研究及应用日益广泛,使得精准、高效的影像学评估成为可能.本文就近年来基于影像图像的AI技术在乳腺病变术前良恶性评估、乳腺癌分类及分级、生物标记物及分子亚型预测、淋巴结病理状态及易感基因诊断等方面的研究现状及进展做一综述,旨在介绍该领域AI发展现状并试图分析当前面临的问题,以期推进乳腺癌AI诊断技术的临床转化,为临床精准无创诊疗提供最佳影像辅助.

Abstract Background: Recent artificial intelligence has exhibited great potential in breast imaging, but its value in precise risk stratification of mammography still needs further investigation. This study is to develop an artificial intelligence system (AIS) for accurate malignancy diagnosis and supportive decision-making on mammographic risk stratification. Methods: In this retrospective study, 49732 mammograms of 24866 breasts from 12815 women from two Asian clinics between August 2012 and December 2018 were included. We developed an AIS using multi-view mammograms and multi-level convolutional neural network features to diagnosis malignancy and further assess the relative strengths of AIS versus current BI-RADS categorization. We further evaluate AIS by conducting a counterbalance-designed AI-assisted study, where ten radiologists read 1302 cases with/without AIS assistance. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, F1 score were measured. Results: The AIS yielded AUC of 0.910 to 0.995 for malignancy diagnosis in the validation and testing sets. Within BI-RADS 3–4 subgroups with pathological results, AIS can downgrade 83.1% of false-positives into benign groups, and upgrade 54.1% of false-negatives into malignant groups. Compared with BI-RADS, AIS performed better sensitivity and specificity in dense and no-calcification subgroups. AIS also can successfully assist radiologists identify 7 out of 43 malignancies initially diagnosed with BI-RADS 0 with specificity of 96.7%. In the counterbalance-designed AI-assisted study, the average AUC across 10 readers was significantly improved with AIS assistance (P = 0.001). Conclusion: AIS can identify malignancy on mammography and further serve as a supportive tool for stratifying BI-RADS categorization.