Development and evaluation of artificial intelligence–assisted dose optimization and image quality enhancement for CT radiography in pediatric patients

Project Abstract

The study presents a comprehensive investigation into AI-assisted dose optimization and image quality enhancement for pediatric CT radiography, addressing the critical balance between diagnostic accuracy and radiation safety in a vulnerable population. We develop a multi-stage framework that integrates data-driven dose estimation, adaptive protocol selection, and image reconstruction improvements to minimize pediatric radiation exposure while preserving, or even enhancing, image quality required for accurate diagnosis. The methodology combines a large, multi-institutional dataset of pediatric CT exams, corresponding dose metrics (CTDIvol, DLP, effective dose), and high-quality ground-truth images obtained under optimized protocols. A supervised deep learning model is trained to predict patient-specific tube current modulation and voxel-wise noise targets from clinical metadata (age, weight, scanned region) and prior imaging, enabling dynamic adjustment of acquisition parameters in real time. Concurrently, a novel AI-enhanced reconstruction pipeline leverages denoising, super-resolution, and artifact suppression techniques tailored for pediatric anatomy to improve low-dose image quality without introducing clinically misleading features. The framework incorporates safety constraints and robust validation against conventional dose-reduction strategies, including iterative reconstruction and scanning protocol harmonization across different scanner platforms. Quantitative evaluation focuses on image quality metrics such as noise power spectrum, contrast-to-noise ratio, structural similarity index, and diagnostic task-based assessments using observer studies with pediatric radiologists. Diagnostic performance is evaluated across common pediatric pathologies, including congenital anomalies, infectious processes, and trauma-related injuries, to ensure clinical relevance. Radiobiological implications are analyzed via Monte Carlo simulations to estimate organ-specific dose reductions and lifetime attributable risk, supplemented by a risk-benefit analysis that accounts for age-specific radiosensitivity. The study also conducts a cost-benefit assessment of implementing the AI framework in clinical workflows, considering training requirements, inference latency, and interoperability with picture archiving and communication systems (PACS) and radiology information systems (RIS). Sensitivity analyses explore model robustness to variations in scanner models, acquisition protocols, and population diversity, with emphasis on generalizability to low-resource settings. An explainability module provides radiologists with interpretable cues about how input features influence dose recommendations and reconstruction decisions, fostering trust and facilitating adoption. The anticipated outcome is a validated, scalable AI-driven solution that achieves clinically acceptable image quality at substantially reduced radiation doses in pediatric CT radiography, along with practical guidelines for integration into routine pediatric imaging workflows, regulatory considerations, and a roadmap for prospective clinical trials. The research contributes to the broader field of radiation protection in medical imaging by demonstrating a synergistic approach that unites dose optimization, image enhancement, and diagnostic accuracy through advanced artificial intelligence, with potential applicability to other pediatric imaging modalities and vulnerable patient populations.

Project Overview

What This Project Is About

---

A plain-language overview of using smart computer tools to adjust CT scan settings for kids and to improve the clarity of the images. The project looks at how artificial intelligence can help reduce the radiation dose a child receives during a scan while keeping the pictures easy to read for doctors.

---

---

The Problem It Addresses

---

Children are more sensitive to radiation, so minimizing dose without sacrificing image quality is important. Traditional CT protocols use one-size-fits-all settings, which can be higher than needed for some children and not ideal for others. This project explores smarter, safer ways to tailor scans.

---

---

Objectives of the Project

---

1. Understand how CT dose relates to image quality in pediatric patients.
2. Explore simple AI ideas that can help adjust scan settings automatically.
3. Test whether the AI-based approach can keep images readable while lowering dose.
4. Identify practical challenges for real-world use in clinics.
5. Suggest guidelines for implementing safer pediatric CT practices.

---

---

What You Will Do Step by Step

---

1. Review basic CT physics and current pediatric dose practices.
2. Learn simple AI concepts that adjust imaging parameters.
3. Simulate data or use anonymized case data to test dose vs. image quality trade-offs.
4. Develop a basic decision tool that suggests dose settings based on patient factors.
5. Evaluate results against standard benchmarks with clear metrics.
6. Analyze limitations and discuss clinical feasibility.
7. Document methods, results, and potential improvements.
8. Prepare a concise report and presentation for stakeholders.

---

---

Expected Outcome

Expected to show that a simple AI-assisted approach can reduce radiation dose in pediatric CT scans while maintaining acceptable image quality, with clear steps for clinical adoption and patient safety benefits.