RESUMEN
BACKGROUND: Accurate tomographic reconstructions require the knowledge of the actual acquisition geometry. Many mobile C-arm CT scanners have poorly reproducible acquisition geometries and thus need acquisition-specific calibration procedures. Most of geometric self-calibration methods based on projection data either need prior information or are limited to the estimation of a low number of geometric calibration parameters. Other self-calibration methods generally use a calibration pattern with known geometry and are hardly implementable in practice for clinical applications. PURPOSE: We present a three-step marker based self-calibration method which does not require the prior knowledge of the calibration pattern and thus enables the use of calibration patterns with arbitrary markers positions. METHODS: The first step of the method aims at detecting the set of markers of the calibration pattern in each projection of the CT scan and is performed using the YOLO (You Only Look Once) Convolutional Neural Network. The projected marker trajectories are then estimated by a sequential projection-wise marker association scheme based on the Linear Assignment Problem which uses Kalman filters to predict the markers 2D positions in the projections. The acquisition geometry is finally estimated from the marker trajectories using the Bundle-adjustment algorithm. RESULTS: The calibration method has been tested on realistic simulated images of the ICRP (International Commission on Radiological Protection) phantom, using calibration patterns with 10 and 20 markers. The backprojection error was used to evaluate the self-calibration method and exhibited sub-millimeter errors. Real images of two human knees with 10 and 30 markers calibration patterns were then used to perform a qualitative evaluation of the method, which showed a remarkable artifacts reduction and bone structures visibility improvement. CONCLUSIONS: The proposed calibration method gave promising results that pave the way to patient-specific geometric self-calibrations in clinics.
Asunto(s)
Procesamiento de Imagen Asistido por Computador , Tomografía Computarizada por Rayos X , Calibración , Procesamiento de Imagen Asistido por Computador/métodos , Fantasmas de Imagen , HumanosRESUMEN
Personalized medicine implies reducing invasiveness of therapeutic procedures. Although interventional radiology proved a very interesting alternative to surgical procedures, it still raises concerns due to the irradiation dose received by the medical team (and by the patient). We propose a novel concept allowing to reduce very significantly the irradiation dose during the phases where tools inserted in the patient have to be tracked with respect to previously acquired images. This implies inserting a miniaturized X-ray detector in the tip of the tools, and reducing the dose by a "rotating collimator". We demonstrate that real-time processing of the signals allows accurate localization of the tip of the tools, with a dose reduction of at least ten times.
Asunto(s)
Cateterismo , Radiología Intervencionista , Interfaz Usuario-Computador , Fluoroscopía , Humanos , Dosis de Radiación , Radiología Intervencionista/instrumentaciónRESUMEN
GATE Monte Carlo simulations are performed to verify sampling schemes in multislice mode PET. Two types of scanner were simulated. The effect of attenuation and scatter were investigated by simulating a realistic phantom. The results are in accordance with the theoretical expectations.