A Multimodal Radiography Image Quality and AGI Framework

Thesis Abstract

The rapid advancement of artificial general intelligence (AGI) and multimodal imaging technologies presents an opportunity to redefine image quality assessment in radiography, addressing diagnostic accuracy, workflow efficiency, and patient safety. Despite gains in single-modality quality metrics, there is a critical gap in integrative frameworks that concurrently optimize spatial resolution, contrast resolution, dose efficiency, and semantic interpretability across diverse imaging modalities (X-ray radiography, CT, and ultrasound) within a single coherent model. This study aims to develop a Multimodal Radiography Image Quality and AGI Framework that synergizes quantitative image quality metrics with clinically meaningful interpretability, enabling AGI-assisted radiographic decision support and standardized quality governance. The specific objectives are (1) to synthesize a theoretical framework linking multimodal image quality determinants with AGI interpretability and clinical decision accuracy; (2) to identify and quantify radiographic image quality metrics across X-ray, CT, and ultrasound modalities using a representative sample; (3) to construct and validate a multimodal image quality index (MIQI) and an AGI interpretability score (AIS) that predict diagnostic confidence and inter-observer agreement; (4) to evaluate the framework’s impact on radiographer workflow, radiation dose efficiency, and patient throughput; and (5) to formulate guidelines for integrating the framework into radiology departments and imaging informatics systems. The methodology adopts a mixed-methods, explanatory sequential design. A purposive sample of 30 hospitals will provide retrospective image datasets comprising 1,500 radiographs, 600 CT slices, and 400 ultrasound clips, along with corresponding radiologist diagnostic reports and blinded image-read auditing data. A convergent parallel quantitative strand will employ regression analyses to model relationships among objective image quality metrics (signal-to-noise ratio, modulation transfer function, contrast-to-noise ratio), dose indicators, and diagnostic accuracy; ANOVA will compare modality-specific quality disparities; and structural equation modeling (SEM) will test the proposed MIQI–AIS latent structure and predictive validity. A qualitative strand will use thematic analysis of semi-structured interviews with 25 radiographers and 15 radiologists to elucidate perceived interpretability, trust, and integration challenges of AGI-assisted assessments. Instrumentation includes standardized image quality assessment tools, dose tracking software, and a bespoke MIQI/AIS measurement instrument developed in alignment with established radiographic quality standards and the International Organization for Standardization (ISO) imaging guidelines. Validity and reliability will be established through content validity indices with expert panels and test–retest procedures, complemented by inter-rater reliability analyses (Cohen’s kappa, intra-class correlation). Data analysis will integrate quantitative results via SEM to validate the theoretical framework, while qualitative findings will be triangulated to contextualize numerical trends and illuminate practical barriers to adoption. The study will draw on relevant theories, including Activity Theory to frame human–AGI–image interactions, and the Technology Acceptance Model (TAM) to interpret adoption likelihood under clinical constraints. Expected findings include a robust MIQI that harmonizes cross-modality image quality facets with a transparent AIS that communicates AGI reasoning steps, leading to improved diagnostic concordance and improved radiographer confidence. It is anticipated that higher MIQI and AIS scores will correlate with increased diagnostic accuracy (p < 0.05) and higher inter-observer agreement. The framework is expected to demonstrate dose efficiency gains without compromising image interpretability, and to reveal workflow benefits such as reduced reporting times and streamlined image review processes. The study aims to contribute to knowledge by delivering a unified, theory-informed framework that operationalizes multimodal image quality within AGI-enabled radiography, offering a validated measurement model, practical integration guidelines, and a decision-support architecture adaptable to diverse clinical settings. The main conclusion will articulate how the Multimodal Radiography Image Quality and AGI Framework can standardize quality governance, enhance clinician trust in AI outputs, and support policy development for imaging informatics. Recommendations will include establishing cross-modality QA protocols, embedding MIQI/AIS dashboards within radiology information systems, investing in user-centered AGI explainability features, and conducting longitudinal studies to assess long-term clinical and operational impacts.