Accurate classification of thyroid nodules is essential for early intervention and improved clinical outcomes; however, its diagnostic performance remains limited by inter-observer variability, limited data availability, and class imbalance in ultrasound imaging. Although existing deep learning models have demonstrated strong performance, they often struggle to capture both global semantic context and fine-grained structural details under limited data. This study introduces a hybrid quantum– classical deep learning framework that leverages quantum variational filtering in tandem with transfer learning via ViT-B16 and EfficientNet-B3 to enhance feature representation. A multi-stage preprocessing pipeline was designed to improve input diversity and class balance. The model jointly performs binary tumor classification and Thyroid Imaging Reporting and Data System (TI-RADS) level prediction, achieving 94.2% accuracy for benign-versus-malignant tumor classification and 91.8% macro-accuracy for TI-RADS prediction. The proposed approach demonstrates promising robustness and generalization in limited-data environments, highlighting the potential of integrating quantum encoding with classical deep learning architectures for medical image diagnostics.
Gamma delta (γδ) T cells exert a pivotal anti-tumor role in the breast cancer (BC) tumor microenvironment, highlighting the importance of investigating their prognostic value for improved patient stratification. We analyzed single-cell ribonucleic acid sequencing data to cluster immune cells and identify marker genes. Prognostic features were selected using Least Absolute Shrinkage and Selection Operator regression in the Cancer Genome Atlas-Breast Invasive Carcinoma cohort and validated across five external Gene Expression Omnibus cohorts. The expression of these prognostic genes was further validated by immunohistochemistry (IHC) in an in-house cohort of BC patients. These selected features were used to construct machine learning models, with the best-performing model undergoing hyperparameter tuning to optimize its performance. γδ T cells were identified as one of the major immune cell populations in BC. Twelve γδ T cell-associated genes were selected based on their prognostic significance in external validation cohorts. The random forest (RF) model achieved the highest accuracy (0.835) after hyperparameter tuning. In external validation, the final model showed the highest performance with area under the curve/accuracy values of 0.81/0.849. IHC analysis confirmed dysregulated expression of key signature proteins in BC tissues. In conclusion, we developed an efficient prognostic RF model based on a 12-gene signature from γδ T cells, which may serve as a clinically valuable tool for risk stratification in BC patients.
Chronic kidney disease is diagnosed from a broad laboratory panel whose size and composition are open to optimization. This study pairs a cooperative hybrid metaheuristic for attribute selection with a weighted voting ensemble. Genetic, swarm, and annealing sub-populations evolved in parallel over one shared fitness landscape, exchanged elites every five iterations, and were refined by annealing applied as a local search on the incumbent; the attributes retained across the elite band formed the subset, which fed four heterogeneous learners whose probabilities were fused using weights tuned from out-of-fold predictions only. Imputation, search, learners, and weights were confined to the training partition of a 10 times five-fold cross-validation, and every single-operator variant, four classical selectors, and the unreduced panel were evaluated under that identical protocol. The ensemble reached 99.00% accuracy, an area under the curve of 0.9994, and an F-measure of 99.19% from 11 of the 24 attributes, dropping 53% of the assays at a cost of two-thirds of a percentage point that is not statistically significant once the dependence between cross-validation folds is modeled. Accuracy held at 98.00% when a further fifth of all values was deleted, and an independent cohort of 200 records recorded as discretized intervals returned 98.00%. The pipeline is statistically equivalent to the strongest single learner and to the strongest classical selector while using fewer than half the attributes.
Publicly accessible conversational artificial intelligence (AI) increasingly informs self-care, yet research among university students has primarily examined awareness, acceptance, and information seeking rather than the progression from AI output to medication action and safety experiences. This cross-sectional study examined 504 University of Jordan students who had previously used AI for self-treatment or self-medication. A 39-item questionnaire assessed symptom consultation, professional verification, medication-related behaviors, and self-reported adverse effects. We derived attitudinal domains using exploratory factor analysis, and modified Poisson regression estimated adjusted prevalence ratios (aPRs). AI was the first source consulted for new symptoms by 36.5% of participants; 42.5% reported purchasing medication following AI advice without consulting a physician or pharmacist (19.0% frequently), 20.0% selected antibiotics following AI-generated information or recommendations, 36.9% reported adverse effects following an AI-recommended medication, and 35.7% modified physician instructions because of conflicting AI advice. Per one-standard-deviation increase, trust/personalization was associated with higher prevalence of all five evaluated outcomes, including any unconsulted purchasing (aPR 1.49, 95% confidence interval 1.33–1.66), whereas risk recognition showed consistent inverse associations (aPR range 0.73–0.80). Physician/pharmacist verification, reported by 64.7%, was associated with lower prevalence of frequent unconsulted purchasing (aPR 0.57, 95% confidence interval 0.40–0.83), reported adverse effects (aPR 0.78, 0.64–0.96), and modification of physician instructions (aPR 0.74, 0.60–0.91), but not with any purchasing or antibiotic selection. Cross-sectional self-report precludes causal attribution. By identifying output appraisal and professional verification as distinct correlates of medication action, this study extends adoption-focused research to clinically relevant behaviors.
Chronic kidney disease (CKD) is a progressive and largely irreversible condition that often remains undetected until advanced stages, and its clinical assessment depends jointly on disease stage, glomerular filtration rate (GFR), and serum creatinine. Most published machine learning models predict only one of these targets and therefore ignore the physiological dependencies linking them. This study introduces an attention-based multi-task learning (ABMTL) framework that predicts all three simultaneously from structured clinical data. A shared representation backbone learns common patterns of renal dysfunction, while three task-specific attention modules re-weight the shared latent features for each objective and feed independent classification and regression heads trained under a joint loss. A leakage-free preprocessing protocol combined correlation-driven adaptive imputation, smoothed target encoding, and training-set-only normalization. The framework was evaluated on 13,500 patient records comprising 13 clinical features, and compared against seven machine learning and deep learning baselines under identical partitions. ABMTL achieved 95.55% accuracy for six-class CKD staging, with coefficients of determination of 0.966 for GFR and 0.976 for serum creatinine, outperforming every baseline on all metrics. Ablation experiments confirmed that the attention modules and multi-task optimization contribute complementary gains, and paired testing indicated the improvements were statistically significant. Kernel Shapley additive explanations analysis identified albumin, blood urea, hemoglobin, hypertension, and diabetes as the dominant predictors, consistent with established nephrology. Serum creatinine and GFR were used only as prediction targets and never as inputs. At the clinically relevant cut-offs of stage ≥3 and stage 4–5, sensitivity exceeded 97% and specificity exceeded 99%. Because the design is single-center, retrospective, and cross-sectional, and calibration and decision-curve analyses have not yet been performed, the framework is presented as a candidate screening-support model that requires external, prospective, and outcome-based validation before clinical use.
