Radiographic dosimetry optimization using AI-driven image quality assessment and adaptive exposure control in computed tomography and fluoroscopy.

Project Abstract

Radiographic dosimetry optimization leveraging AI-driven image quality assessment (IQA) and adaptive exposure control (AEC) across computed tomography (CT) and fluoroscopy to reduce patient radiation dose while preserving diagnostic performance. This study integrates deep learning-based IQA models with physics-informed exposure modulation to dynamically adjust kVp, mA, and pulse sequences in real time, informed by task-specific image quality targets and modality-specific constraints. The research develops a unified framework that (i) characterizes the relationship between radiation dose metrics (CTDIvol, DLP, fluoroscopy air kerma) and image quality indicators (SNR, CNR, spatial resolution, noise power spectrum) under varying patient sizes and clinical tasks; (ii) designs robust AI-driven IQA predictors trained on large, annotated radiographic datasets to estimate diagnostically relevant image quality from raw acquisition data and reconstructed images; (iii) implements an adaptive exposure controller that optimizes acquisition parameters using reinforcement learning and constraint-based optimization to meet predefined clinical thresholds while minimizing dose; (iv) incorporates domain knowledge from radiation physics, scatter modeling, and detector characteristics to maintain dose efficiency across protocols, including contrast-enhanced CT, low-dose CT for screening, interventional fluoroscopy, and performance-guided fluoroscopic angio. A multi-modal dataset comprising phantom studies, retrospective clinical scans, and prospective workflow data will be compiled to train and validate the AI models, with cross-institutional external validation to ensure generalizability. The methodology includes (a) data curation and preprocessing, (b) development of IQA metrics aligned with radiologist judgment and diagnostic tasks, (c) design of an end-to-end AI-enabled dose optimization engine integrating parameter estimation, dose tracking, and real-time feedback, (d) simulation-based evaluation using digital phantoms and Monte Carlo dose calculations, and (e) prospective pilot testing in clinical workflows to assess dose reduction, image quality consistency, diagnostic confidence, and throughput impact. Expected outcomes include statistically significant reductions in effective dose and organ doses without compromising lesion detectability or diagnostic accuracy, demonstrated through receiver operating characteristic analysis, task-based performance studies, and reader studies. The study also investigates model interpretability, robustness to noise, data drift, and variability in scanner hardware, with an emphasis on incorporating safety constraints to prevent under-dosing. The research contributes a scalable, transfer-ready blueprint for AI-assisted dosimetry management in radiography, offering guidelines for regulatory compliance, quality assurance, and integration with existing picture archiving and communication systems (PACS) and radiology information systems (RIS). Ultimately, the project aims to establish a clinically viable paradigm where adaptive, AI-informed exposure modulation achieves substantial radiation dose reduction while sustaining high diagnostic fidelity across CT and fluoroscopic procedures.

Project Overview

What This Project Is About

---

A plain-language overview of the topic and what the project investigates.

---

---

The Problem It Addresses

---

Explain a gap in current radiography practice: balancing image quality with patient dose in CT and fluoroscopy, and how AI might help adjust exposure in real time.

---

---

Objectives of the Project

---

1. Assess how AI tools can evaluate image quality during radiographic procedures.
2. Develop a simple adaptive exposure approach that uses AI feedback to optimize dose.
3. Test the approach on CT and fluoroscopy datasets to compare image quality and dose.
4. Provide guidelines for safe and effective implementation in clinical settings.

---

---

What You Will Do Step by Step

---

1. Review basic radiography concepts and relevant AI image assessment methods.
2. Collect or simulate paired image-quality and dose data from CT and fluoroscopy cases.
3. Train a lightweight AI model to predict optimal exposure settings based on current image signals.
4. Integrate a decision framework that adapts exposure during scanning or imaging sessions.
5. Evaluate performance by comparing image quality, diagnostic adequacy, and dose metrics against standard practice.
6. Discuss practical considerations and limitations for clinical use.

---

---

Expected Outcome

Anticipate a validated method that maintains or improves image quality while reducing radiation dose, with clear recommendations for implementation and future research.