Multimodal AI and Process Analytical Technology for Solid-Dosage Quality Inspection: A Narrative Review

Background/Objectives: Automated inspection is valuable when visual and process signals map to critical quality attributes (CQAs). This narrative review evaluates computer vision and complementary process analytical technology (PAT) for surface integrity, coating structure, chemical distribution, internal defects, and release-relevant quality. Methods: We synthesize selected literature on solid-dosage inspection, multimodal sensing, machine learning, and hybrid digital twins using a targeted narrative approach. Results: Visible imaging supports rapid surface inspection; optical coherence tomography and terahertz methods resolve coating and subsurface structure; Raman, near-infrared, and hyperspectral methods add chemical or spatial evidence. Task-specific pharmaceutical studies report promising performance, but rare defects, data leakage, domain shift, calibration, missing inputs, throughput, and lifecycle drift constrain generalization. CQA-driven fusion requires product-specific validation, calibrated uncertainty, human review, change control, and traceable data. Conclusions: Hybrid digital twins can support prospective scenario analysis and process understanding, but current evidence does not support autonomous batch release. Inspection outputs remain in-process evidence unless validated within a control strategy or surrogate relationship for the specified release attribute.