Three-dimensional (3D) bioprinting of hydrogel inks is an enabling technology for tissue engineering and the fabrication of personalized drug-delivery systems. However, in most laboratories the extrusion parameters are still optimized by slow, resource-intensive trial-and-error, because printability depends on the combined and often nonlinear influence of ink composition, extrusion pressure, and nozzle geometry rather than on any single variable. Alginate–methylcellulose (Alg–MC) is a widely used pharmaceutical bioink whose shear-thinning behavior makes this optimization particularly demanding. Here we developed a physics-informed machine-learning framework that predicts extrusion print quality before printing. A hybrid dataset of 3280 labeled samples was assembled by combining laboratory extrusion experiments, computational fluid dynamics (CFD) simulations of nozzle flow, and controlled augmentation with a deep belief network (DBN). Five physics-informed features encoding pressure-to-diameter and shear-rate interactions were engineered to embed rheological domain knowledge. Five base classifiers alongside two ensemble strategies, seven models in all, were benchmarked under five-fold stratified cross-validation. The voting ensemble performed best (accuracy, 93.3%; F1 = 0.929), narrowly ahead of the stacking ensemble (93.3% accuracy; F1 = 0.928). Alg:MC ratio, extrusion pressure, and the pressure–diameter interaction term stood out as the dominant predictors, consistent with established alginate rheology. On external validation—withholding one composition ratio and one needle gauge entirely from training and testing on those genuinely unseen conditions—accuracy was 75.9%, a stricter test than a random held-out split. Overall, the framework offers a practical, interpretable way to cut down experimental burden in pharmaceutical bioink development.

Medical image classification models often struggle to generalise across heterogeneous clinical domains owing to variations in visual characteristics, acquisition conditions, and class imbalance. Existing studies largely address these challenges independently, with limited investigation into their combined impact on classification robustness. This paper proposes a lightweight dual-stream framework based on a pretrained ConvNeXt-Tiny backbone, integrating complementary global semantic and local structural feature representations with imbalance-aware optimisation. Rather than modifying the backbone, the framework enhances feature discrimination through dual-stream processing while incorporating class-balanced focal loss and stratified sampling for minority-class recognition. The framework is evaluated independently on four public datasets spanning dermatology, ophthalmology, and gastrointestinal endoscopy using five-fold cross-validation to assess architectural robustness across heterogeneous domains rather than cross-domain transfer of a single trained model. Experimental results show strong performance across all datasets, achieving AUC values above 0.92. Ablation studies confirm that dual-stream representation learning provides the primary performance gains, while imbalance-aware optimisation further improves robustness under severe class imbalance. These findings demonstrate an effective and practical solution for medical image classification across heterogeneous clinical imaging domains without requiring dataset-specific architectural modifications. Limitations regarding independent per-domain (rather than cross-domain) evaluation, baseline comparability, and incomplete quantitative calibration analysis are discussed explicitly and identified as directions for future work.