Guanqiao Jin

Background: Multidrug-resistant tuberculosis (MDR-TB) poses an escalating public health challenge that complicates diagnosis and treatment. Early detection is crucial for improving the outcomes. This study aimed to evaluate the diagnostic performance of whole-lung computed tomography (CT) radiomics features combined with clinical characteristics in distinguishing MDR-TB from drug-sensitive tuberculosis (DS-TB). Methods: This retrospective study included 750 patients with MDR-TB and DS-TB from two hospitals. Clinical data and non-contrast CT images were obtained. The radiomic features were extracted using PyRadiomics. A three-step feature selection process, including t-tests/U-tests, Pearson correlation, and the least absolute shrinkage and selection operator (LASSO), was employed to identify the optimal features. Diagnostic models based on clinical and radiomic features were constructed using LightGBM and multilayer perceptron (MLP) algorithms, respectively. A combined model integrated both types of features. Model performance was assessed using area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 score. Results: Diabetes mellitus and tuberculosis (TB) retreatment were identified as independent risk factors for MDR-TB. The clinical model achieved AUC values of 0.742, 0.738, and 0.725 for training, internal validation, and external validation sets, respectively. Seven radiomics features were selected, with the radiomics model achieving AUC values of 0.724, 0.720, and 0.703. The combined model outperformed the individual models, with AUC values of 0.816, 0.795, and 0.835, and superior sensitivity and specificity. Conclusions: Integrating whole-lung CT radiomics with clinical features significantly enhances the diagnostic accuracy of MDR-TB. The combined model outperforms individual models, underscoring the potential of radiomic-clinical data integration. This approach could expand MDR-TB screening coverage without additional economic burden, thereby facilitating prevention and control.

Background: Pulmonary nodules categorized as Lung Imaging Reporting and Data System (Lung-RADS) 3 and 4A constitute a substantial proportion of radiologically indeterminate lesions, and current imaging criteria demonstrate limited accuracy for distinguishing benign from malignant nodules. To develop and validate an integrated predictive model combining radiomics and two-and-a-half (2.5D) deep transfer learning (DTL) for differentiating benign from malignant pulmonary nodules classified as Lung-RADS categories 3 and 4A on computed tomography (CT). Methods: This retrospective study included 298 patients with Lung-RADS 3 and 4A nodules from three centers. The cohort from Center 1 (n=247) was divided into training (n=172) and test (n=75) sets, while patients from Centers 2 and 3 (n=51) formed an independent validation set. We constructed three models: a radiomics (Rad) model, a DTL model, and an integrated deep transfer radiomics (DTR) model. Model performance was evaluated using the area under the receiver operating characteristic curve, and clinical utility was assessed using decision curve analysis. Results: The DTR model demonstrated superior performance in the training [area under the curve (AUC): 0.975, 95% confidence interval (CI): 0.9529–0.9981], testing (AUC: 0.851, 95% CI: 0.7328–0.9683), and external validation cohorts (AUC: 0.727, 95% CI: 0.5876–0.8663), outperforming both the Rad model (training: AUC: 0.743; testing: AUC: 0.642; validation: AUC: 0.613) and the DTL model (training: AUC: 0.843; testing: AUC: 0.757; validation: AUC: 0.701). The DTR model exhibited high specificity (0.895) and positive predictive value (0.956) in the test cohort. SHapley Additive exPlanations (SHAP) analysis revealed that the DTR model effectively leveraged complementary features from both Rad and DTL. Conclusions: The integration of Rad and 2.5D DTL significantly improves the diagnostic accuracy for differentiating between benign and malignant Lung-RADS 3 and 4A nodules. This approach provides a robust decision-support tool that could potentially reduce unnecessary interventions for benign nodules while facilitating earlier detection and treatment of malignant lesions.

Nasopharyngeal carcinoma（NPC） is a distinct type of head and neck cancer closely associated with Epstein-Barr virus（EBV） infection and exhibits significant geographic variations in its incidence. Despite recent advancements in radiotherapy techniques and precision medicine for NPC, the overall survival rate remains unsatisfactory due to tumor metastasis, recurrence, and drug resistance. Single-cell RNA sequencing（scRNA-seq） is an emerging technology that allows for the analysis of gene expression at single-cell resolution, providing a clearer understanding of tumor cell subpopulations, the evolutionary trajectory of tumor cells, and the functional roles and interactions of cells within the tumor microenvironment. This provides new ideas for the development of precision medicine in NPC. Here, we review the applications of scRNA-seq in exploring the mechanisms of NPC pathogenesis, tumor heterogeneity, the tumor microenvironment, drug resistance, and therapeutic response.

PD-1/PD-L1 immunotherapy represents the primary treatment for advanced NSCLC patients; however, response rates to this therapy vary among individuals. This dual-center study aimed to integrate habitat radiomics and multi-instance deep learning to predict durable clinical benefits from immunotherapy. We retrospectively collected 590 NSCLC patients from two medical centers who received PD-1/PD-L1 inhibitor immunotherapy. Patients from the GMU center were divided into a training cohort (n = 375) and an internal validation cohort (n = 161) for habitat analysis and multi-instance deep learning model development. Patients from the YJ center formed an external testing cohort (n = 54) for model validation. We implemented a DenseNet121-based architecture extracting radiomics features from triplanar (axial/coronal/sagittal) tumor sequences to construct a 2.5D deep-learning dataset. Then, we fuse 2.5D features through multi-instance learning. Additionally, we use K-means clustering to divide the tumor VOI into three subregions to extract radiological features for building a Habitat model. Finally, we use the Extra-Trees classifier to construct MIL, Habitat, and Combined models, the Combined model integrating age factors into the analysis. The primary endpoint was durable clinical benefit. Finally, a separate PD-L1 expression dataset was used to compare the predictive performance of imaging models against PD-L1 status (positive/negative) and expression levels (high/low) to identify the optimal model for predicting immunotherapy clinical benefit. The Combined model combining Habitat, MIL, and patient age demonstrated robust DCB prediction with AUCs of 0.906(95% CI: 0.874–0.936), 0.889(95% CI: 0.826–0.948), and 0.831 (95% CI: 0.710–0.927)in training, validation, and testing cohorts respectively. Comparative analysis revealed all imaging models outperformed PD-L1 expression status (positive/negative) and levels (high/low) in predicting therapeutic response, with Habitat analysis showing superior performance to MIL alone. Notably, peritumoral structural features emerged as significant predictors of treatment efficacy. This non-invasive predictive framework provides clinically actionable insights for immunotherapy stratification, potentially overcoming limitations of current biomarker testing while highlighting the prognostic value of spatial tumor heterogeneity analysis.

Occult lymph node metastasis (OLNM) refers to lymph node involvement that remains undetectable by conventional imaging techniques, posing a significant challenge in the accurate staging of lung adenocarcinoma. This study aims to investigate the potential of combining 2.5D deep learning radiomics with clinical data to predict OLNM in lung adenocarcinoma. Retrospective contrast-enhanced CT images were collected from 1,099 patients diagnosed with lung adenocarcinoma across two centers. Multivariable analysis was performed to identify independent clinical risk factors for constructing clinical signatures. Radiomics features were extracted from the enhanced CT images to develop radiomics signatures. A 2.5D deep learning approach was used to extract deep learning features from the images, which were then aggregated using multi-instance learning (MIL) to construct MIL signatures. Deep learning radiomics (DLRad) signatures were developed by integrating the deep learning features with radiomic features. These were subsequently combined with clinical features to form the combined signatures. The performance of the resulting signatures was evaluated using the area under the curve (AUC). The clinical model achieved AUCs of 0.903, 0.866, and 0.785 in the training, validation, and external test cohorts The radiomics model yielded AUCs of 0.865, 0.892, and 0.796 in the training, validation, and external test cohorts. The MIL model demonstrated AUCs of 0.903, 0.900, and 0.852 in the training, validation, and external test cohorts, respectively. The DLRad model showed AUCs of 0.910, 0.908, and 0.875 in the training, validation, and external test cohorts. Notably, the combined model consistently outperformed all other models, achieving AUCs of 0.940, 0.923, and 0.898 in the training, validation, and external test cohorts. The integration of 2.5D deep learning radiomics with clinical data demonstrates strong capability for OLNM in lung adenocarcinoma, potentially aiding clinicians in developing more personalized treatment strategies.

This study aims to develop an integrated model combining habitat-based radiomics and clinical data to predict lymph node metastasis in patients with clinical N0 peripheral lung adenocarcinomas measuring ≤ 3 cm in diameter. We retrospectively analyzed 1132 patients with lung adenocarcinoma from two centers who underwent surgical resection with lymph node dissection and had preoperative computed tomography (CT) scans showing peripheral nodules ≤ 3 cm. Multivariable logistic regression was employed to identify independent risk factors for the clinical model. Radiomics and habitat models were constructed by extracting and analyzing radiomic features and habitat regions from contrast-enhanced CT images. Subsequently, a combined model was developed by integrating habitat-based radiomic features with clinical characteristics. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). The habitat model exhibited promising predictive performance for lymph node metastasis, outperforming other standalone models with AUCs of 0.962, 0.865, and 0.853 in the training, validation, and external test cohorts, respectively. The combined model demonstrated superior discriminative ability, achieving the highest AUCs of 0.983, 0.950, and 0.877 for the training, validation, and external test cohorts, respectively. The integration of habitat-based radiomic features with clinical data offers a non-invasive approach to assess the risk of lymph node metastasis, potentially supporting clinicians in optimizing patient management decisions.

The optimal surgical strategy for lymph node dissection in lung adenocarcinoma remains controversial. Accurate predicting occult lymph node metastasis (OLNM) in patients with clinical T1 lung adenocarcinoma is essential for optimizing treatment decisions and improving patient outcomes. This study analyzes the relationship between anaplastic lymphoma kinase (ALK) status, clinicopathological characteristics, computed tomography (CT) features, and OLNM in patients with clinical T1 lung adenocarcinoma. A retrospective analysis was conducted on data from patients with clinical T1 lung adenocarcinoma who showed no lymph node metastasis on preoperative CT and underwent surgical resection with lymph node dissection at two centers from January 2016 to December 2023. Univariate and multivariate logistic regression analyses were performed to identify factors associated with OLNM. Among 1138 patients with clinical T1 lung adenocarcinoma, 167 (14.6%) were found to have OLNM, including 55 (4.8%) with pathological N1 status and 112 (9.8%) with pathological N2 status. Multivariate logistic regression analysis identified lobulation, spiculation, solid density, lymphovascular invasion, spread through air spaces (STAS), micropapillary pattern, solid pattern, and carcinoembryonic antigen (CEA) levels as independent positive predictors of OLNM. Furthermore, lobulation, lymphovascular invasion, STAS, micropapillary pattern, solid pattern, CEA levels, and ALK were independent positive predictors of occult N2 lymph node metastasis. The lepidic pattern, however, was identified as an independent negative predictor for OLNM and occult N2 lymph node metastasis. The identified predictors may assist clinicians in evaluating the risk of OLNM in patients with clinical T1 lung adenocarcinoma, potentially guiding more targeted intervention strategies.