Regulatory Science for Engineering Intuitive, Engaging, Safe and Effective Human-Device Interaction
(Bahaa Ghammraoui, FDA)
Background: Photon-counting CT (PCCT) enables multi-energy imaging and material decomposition (e.g., iodine mapping), but quantitative performance can be sensitive to detector non-idealities. Pixel-to-pixel variations such as energy threshold shifts and miscalibrated detector elements can lead to ring artifacts that propagate through reconstruction and may bias spectral outputs. While many correction methods focus on visual improvement, it is important to quantify whether artifact removal changes clinically relevant endpoints such as detectability of low-contrast objects and accuracy of iodine estimation. This project leverages physics-based PCCT simulation tools to generate realistic CT data with controlled detector miscalibration and to evaluate AI/ML-based correction approaches using task-based and quantitative metrics.
Research Plan: The aim of this study is to develop a simulation-driven framework to train and evaluate AI/ML ring artifact correction and quantify its impact on spectral CT performance. The student will use existing simulation software to create controlled pixel threshold variation and miscalibrated pixel patterns to generate paired “artifact-corrupted” and “reference/clean” datasets using anthropomorphic phantoms. A neural network will be trained to reduce ring artifacts in raw data (or per-energy reconstructed images, as appropriate to the pipeline) while preserving spectral consistency. The trained model will then be evaluated on a uniform phantom with iodine inserts of varying size and concentration to assess (1) detectability-related metrics and (2) iodine estimation accuracy and precision from iodine mapping/material decomposition. For a 3‑month internship, the primary deliverable is a well-documented dataset generation workflow and training dataset; evaluation testing will be a plus if time allows.
Prerequisites: Engineering or science coursework in math and physics; numerical/scientific computing; 1+ year programming in Python. Familiarity with CT imaging, image reconstruction, or machine learning is a plus.
Bahaa Ghammraoui is a Medical Imaging Research Scientist and Regulatory Science Reviewer at the U.S. Food and Drug Administration (FDA), Center for Devices and Radiological Health (CDRH). His research focuses on advanced X-ray imaging technologies, including photon-counting detectors, spectral CT, phase-contrast imaging, and coherent scatter imaging. His expertise includes imaging system design, detector performance, and quantitative image quality evaluation. He conducts regulatory reviews of medical imaging devices, including CT systems, photon-counting detector–based technologies, and AI/ML-enabled imaging software. He is also involved in developing and sharing open-source physics-based simulation tools for photon-counting CT and other advanced X-ray imaging modalities through the FDA’s DIDSR GitHub. In addition, he is a professor at the University of Maryland, where he teaches Python programming, imaging physics, and image processing.