Find the papers that actually matter

Although artificial intelligence, including machine learning offers transformative potential, its use in infectious disease epidemiology has been understudied. Emerging infectious diseases including WNV infection remains a leading cause of mortality globally. The study aimed to provide a decade-long insight into the global trends on WNV outbreaks. The study offers a ML-powered framework to guide preventive strategies. Environmental indices including neighbourhood greenness, forest cover, and temperature were captured using 250m resolution MODIS satellite imagery. Seasonal human, avian and equine WNV surveillance records and other regional indicators for Italy from 2013 to 2023 were retrieved from surveillance repositories. Spatial trends and interactions were evaluated using spatial autocorrelation analyses. Further, a ML architecture was developed for prediction of high or low WNV classification based on satellite imagery of climatic indices (deep neural networks) compared with non-raster variables (gradient boosting machine). Moran's I estimation, confirmed by the Monte Carlo approximation indicated evidence of significant spatial autocorrelation with Moran's I values of 0.0561 (P < 0.00259) and 0.0561 (P < 0.009) respectively while significant clustering was observed in regions of northern Italy. The ML-based deep neural network yielded an accuracy of 80% and 79.7% for NDVI- and LST-based raster inputs, while GBM model isolated key predictors of human WNV confirmed cases with an AUC of 0.835. This study presents a scalable approach for human WNV outbreak prediction incorporating geospatial analyses and artificial intelligence, providing insights on the disease epidemiology and a framework for other emerging infectious disease investigations.

Cardiovascular disease (CVD) is the leading cause of death globally [1] as well as the leading cause of death among cancer survivors [2]. The outcomes of CVD mortality among cancer patients, particularly those with breast cancer, highlight the need for early detection of CVD at the beginning of cancer treatment as cardiotoxicity can also lead to accelerated development of chronic diseases, especially in the presence of risk factors [3]. A fusion deep learning model was developed and tested using computed tomography (CT) scans and electronic health records (EHR) for CVD mortality prediction in breast cancer patients undergoing radiation therapy. The model utilizes computed tomography (CT) scans and electronic health records (EHR) for CVD mortality prediction in breast cancer patients undergoing radiation therapy. A cohort of 23,067 patients consisting of ∼5 million CT slices and ∼600,000 EHR documents was used for the model development and testing. Performance of the model is assessed using the AUC and accuracy at a 95% confidence level. The fusion model achieves an AUC of 0.946 [0.939---0.950], and accuracy of, 0.93 [0.92 - 0.94] at 95% confidence interval (CI). These results show that fusion models can learn versatile representations from medical images and medical text documents and can effectively be combined for tasks like predicting CVD mortality with higher accuracy when employing the appropriate fusion strategy.

Accurate survival prediction is a key component to patient quality of life (QoL)-centered treatment. It allows, for example, setting reasonable treatment goals or timely palliative care referral. However, clinical estimates are often too optimistic leading to lower patient QoL before death. Patient-reported outcomes (PROs) have proven to be an important survival predictor improving the overall prognostic accuracy. In this study, transformer architecture was explored to leverage PRO trajectories to improve survival accuracy and identify the most prognostic PRO symptoms for survival prediction. We analyzed 475 (cross-validation discovery set: 380; held-out testing set: 95) early-stage non-small cell lung cancer (NSCLC) patients who underwent SBRT treatment and routinely completed the Edmonton Symptom Assessment Scale (ESAS). A transformer-based model was developed to perform longitudinal modeling of overall survival (OS), incorporating PROs collected at multiple post-treatment follow-ups, as well as clinical and demographic variables. The performance of the proposed model was compared to traditional outcome modeling approaches, including univariate and multivariate (time-varying) Cox proportional hazards regression (CoxPH) and joint probability survival modelling. The best-performing transformer model was interpreted using the SHapley Additive exPlanation (SHAP) values, and the most prognostically relevant ESAS symptoms were identified through a backward elimination procedure guided by concordance index (c-index) and area under the ROC curve (AUC). The best-performing transformer model achieved a cross-validated c-index of 0.753 [95 % CI: 0.742-0.764] and an AUC of 0.862 [95 % CI: 0.846-0.878] on the discovery set. On the heldout test set, the model reached a c-index of 0.694 and an AUC of 0.785, evaluated at the last time point. It significantly outperformed both Cox models and the joint probability model. Model interpretation using SHAP values and backward elimination identified appetite loss, pain, overall well-being, and shortness of breath as the most prognostically relevant symptoms for survival prediction. Transformer-based survival models that integrate longitudinal PROs significantly enhance prognostic accuracy in SBRT-treated NSCLC patients. Loss of appetite and pain emerged as the most predictive symptoms, followed by overall wellbeing and shortness of breath. These findings suggest that targeted, symptom-focused PROs tracking could streamline clinical implementation and improve survival estimation in routine oncology care.

CRISPR-Cas12a has become a versatile biotechnology platform with important applications in biosensing, diagnostics, and precision genome editing. This system is activated by a single crRNA, exhibits target-responsive trans-cleavage activity, and recognizes programmable PAM sequences. These features provide a robust basis for accurate detection of diverse biomarkers. Its detection capabilities include nucleic acid targets such as viral RNA and cancer mutations, as well as non-nucleic acid molecules like exosomes and proteins. Recent advancements have shown significant advantages, including multi-temperature adaptability, rapid kinetics, and compatibility with both DNA and RNA targets. Technical improvements include machine learning-assisted crRNA design for enhanced prediction accuracy and engineered EnAsCas12a variants that overcome conventional PAM restrictions. Notable achievements involve entropy-driven circuits that achieve attomolar-level sensitivity, smartphone-compatible four-channel quantitative detection systems, and streamlined integrated workflows completed within 30 min. Advances in sensor design, such as metal-organic framework encapsulation and high-performance aptamer-based sensors, have further expanded detection capabilities. In oncology research, CRISPR-Cas12a technology provides powerful tools to comprehensively analyze complex molecular networks within the tumor microenvironment (TME) and facilitate ultrasensitive detection of early-stage cancer biomarkers. Additionally, in genome editing, CRISPR-Cas12a enables precise genomic modifications due to distinct repair pathways, versatile delivery methods, and efficient creation of transgenic models. Thus, it expands its functional scope beyond diagnostics. With ongoing development, this technology is expected to evolve into an integrated platform combining TME research, point-of-care cancer diagnostics, and programmable genome engineering, offering innovative solutions for both biomedical research and clinical translation.

Histopathological image analysis is critical for cancer diagnosis, yet many existing models suffer from limited interpretability, high computational demands, and suboptimal classification accuracy. To overcome these limitations, we propose a novel model, Complementary Residual Retentive Network with Guided Gaussian Combined Arms Algorithm (C2RN2GC2A), designed to enhance efficiency and accuracy in cancer classification from histopathological images. C2RN2GC2A is a deep learning model that assimilates residual learning with optimized Gaussian perturbations, thus enhancing both feature extraction and working time in classification tasks. The system merges 2GC2A, a metaheuristic optimization approach motivated by military tactics, for the purpose of improving feature selection, truncating training loss, and speeding up convergence. The Two-stage Guided Chaotic Capuchin Algorithm (2GC2A) brings together Gaussian perturbations with a combined arms tactic, which makes it possible to do a good job of both exploring and exploiting the search space for better parameter tuning. In order to achieve interpretability, Layer-wise DeepLIFT Relevance Propagation (LDLRP) is used to delineate the significant areas of the image that have an impact on the classification, thus making the process more transparent and building up clinical trust. LDLRP is a cutting-edge explainable AI technology that grants relevance ratings to input characteristics and thus allows the model to visually demonstrate the most significant regions in histopathological images, thereby facilitating clinical decision-making. Testing on LC25000 produced a remarkable accuracy of 98.02% along with a minuscule training loss of 0.08, besides which there were 13 false positives and 29 false negatives. On BreakHis, the accuracy was 98.54%, and the validation loss was 0.05, with 98 false positives and 112 false negatives. The proposed framework significantly improves diagnostic reliability, classification accuracy, and clinical transparency in multi-cancer histopathological image analysis.

In the phase 3 KEYNOTE-042 China study of participants enrolled in China in the global KEYNOTE-042 (NCT02220894) and China extension (NCT03850444) studies, pembrolizumab improved overall survival (OS) versus chemotherapy in locally advanced or metastatic non-small-cell lung cancer (NSCLC) with programmed cell death ligand 1 (PD-L1) tumor proportion score (TPS) ≥50% (hazard ratio [HR], 0.63; 95% CI, 0.43-0.94), ≥20% (0.66; 0.47-0.92), and ≥1% (0.67; 0.50-0.89). We present outcomes from this study after 5 years of follow-up. Chinese participants with previously untreated locally advanced or metastatic NSCLC with PD-L1 TPS ≥1% without EGFR or ALK alterations were eligible. Participants were randomized 1:1 to pembrolizumab 200 mg every 3 weeks for up to 35 cycles or carboplatin plus paclitaxel or pemetrexed with optional pemetrexed maintenance (nonsquamous only). Primary endpoints were OS in the PD-L1 TPS ≥50%, ≥20%, and ≥1% subgroups. Median follow-up was 63.7 (range, 56.3-72.6) months among 262 participants (pembrolizumab, n = 128; chemotherapy, n = 134) included in this study. Pembrolizumab prolonged OS versus chemotherapy in participants with PD-L1 TPS ≥50% (HR, 0.65; 95% CI, 0.45-0.93), ≥20% (0.67; 0.49-0.91), and ≥1% (0.66; 0.51-0.87). Grade 3 to 5 treatment-related AEs occurred in 19.5% and 68.8% of participants in the pembrolizumab and chemotherapy groups, respectively. In conclusion, after 5 years of follow-up, pembrolizumab continued to demonstrate improved OS versus chemotherapy with manageable safety in Chinese participants with previously untreated locally advanced or metastatic NSCLC that expressed PD-L1. These data further support pembrolizumab monotherapy as a standard of care for these patients.

Leveraging multimodal information from Magnetic Resonance Imaging (MRI) plays a vital role in lesion segmentation, especially for brain tumors. However, in clinical practice, multimodal MRI data are often incomplete, making it challenging to fully utilize the available information. Therefore, maximizing the utilization of this incomplete multimodal information presents a crucial research challenge. We present a novel meta-guided multi-modal learning (MGML) framework that comprises two components: meta-parameterized adaptive modality fusion and consistency regularization module. The meta-parameterized adaptive modality fusion (Meta-AMF) enables the model to effectively integrate information from multiple modalities under varying input conditions. By generating adaptive soft-label supervision signals based on the available modalities, Meta-AMF explicitly promotes more coherent multimodal fusion. In addition, the consistency regularization module enhances segmentation performance and implicitly reinforces the robustness and generalization of the overall framework. Notably, our approach does not alter the original model architecture and can be conveniently integrated into the training pipeline for end-to-end model optimization. We conducted extensive experiments on the public BraTS2020 and BraTS2023 datasets. Compared to multiple state-of-the-art methods from previous years, our method achieved superior performance. On BraTS2020, for the average Dice scores across fifteen missing modality combinations, building upon the baseline, our method obtained scores of 87.55, 79.36, and 62.67 for the whole tumor (WT), the tumor core (TC), and the enhancing tumor (ET), respectively. We have made our source code publicly available at https://github.com/worldlikerr/MGML.

Breast cancer remains the leading cause of cancer-related mortality among women worldwide. Early screening and accurate subtype classification are critical for guiding clinical decision-making. In this study, we present MGScreener, an interpretable multi-view framework that integrates dual-view mammography images, including cranial-caudal (CC) and mediolateral oblique (MLO) views, with patient-level clinical data. The model incorporates an active learning strategy to address two key classification tasks: distinguishing benign from malignant breast lesions and identifying invasive ductal carcinoma (IDC) among malignant subtypes. An entropy-based uncertainty sampling method is employed to select highly informative cases from 210 unlabeled samples for prioritized annotation, substantially reducing manual labeling costs. Across two independent test sets, MGScreener-1 achieved an accuracy of 89.7 % and an ROC-AUC of 0.941 for benign-versus-malignant classification (Task 1), while MGScreener-2 achieved an accuracy of 88.2 % and an ROC-AUC of 0.884 for IDC identification (Task 2). With MGScreener assistance, radiologists improved their diagnostic accuracy by more than 10 % compared with independent reading. Overall, MGScreener offers a scalable and effective solution for precise breast cancer screening and molecular subtype classification.

There are concerns about the fairness of clinical prediction models. "Fair" models are defined as those for which their performance or predictions are not inappropriately influenced by protected attributes such as ethnicity, gender, or socioeconomic status. Researchers have raised concerns that current algorithmic fairness paradigms enforce strict egalitarianism in healthcare, leveling down the performance of models in higher-performing subgroups instead of improving it in lower-performing ones. We propose assessing the fairness of a prediction model by expanding the concept of net benefit, using it to quantify and compare the clinical impact of a model in different subgroups. We use this to explore how a model distributes benefits across a population, its impact on health inequalities, and its role in the achievement of health equity. We show how resource constraints might introduce necessary trade-offs between health equity and other objectives of healthcare systems. We showcase our proposed approach with the development of two clinical prediction models: (1) a prognostic type 2 diabetes model used by clinicians to enroll patients into a preventive care lifestyle intervention programme and (2) a lung cancer screening algorithm used to allocate diagnostic scans across the population. This approach helps modelers better understand if a model upholds health equity by considering its performance in a clinical and social context.

To predict cartilage tumor malignancy from radiographic images combined with readily available non-imaging information based on a vision-language foundation model. This single-institution study assembled a dataset of 3336 radiographs from 455 patients with enchondroma or chondrosarcoma that was assembled from two sources: (1) patients with histopathology-confirmed diagnoses of enchondroma or chondrosarcoma, and (2) patients with imaging-stable enchondroma without biopsy, confirmed through long-term imaging follow-up. An adapted vision-language foundation model based on the pre-trained CLIP (Contrastive Language-Image Pretraining) architecture was fine-tuned with our proposed Medical Knowledge Adapters and evaluated using 10-fold patient-level cross-validation to predict cartilage tumor malignancy from plain radiographs and demographic information. Using radiographs alone, the model achieved an Areas Under the receiver operating characteristic Curve (AUC) of 0.91 ± 0.04. Incorporating demographics improved the AUC to 0.94 ± 0.02. Subgroup analysis demonstrated robust generalizability across tumor grades with an AUC of 0.91 ± 0.07 in distinguishing atypical cartilaginous tumors (ACT) previously known as low grade chondrosarcomas, and 0.95 ± 0.02 in differentiating high-grade chondrosarcomas from enchondromas. Within the clinically challenging extremity subgroup (enchondroma vs ACT/LGCS), the model achieved an AUC of 0.79 ± 0.14, reflecting diagnostic difficulty observed in clinical practice. This foundation model-based approach demonstrates strong performance using accessible data sources, offering a non-invasive, cost-efficient, and scalable solution for cartilage tumor assessment in musculoskeletal oncology.