High-precision position estimation in PET using artificial neural networks

Mateo, Fernando; Aliaga, Ramón J.; Ferrando, N.; Martínez, J.D.; Herrero, V.; Lerche, Christoph; Colom, R.J.; Monzó, J.M.; Sebastiá, A.; Gadea, R.

doi:10.1016/j.nima.2009.01.058

Cited by 12 publications

(10 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In recent publications, the successful application of Neural Networks for γ-ray positioning in PET was reported by several groups [27,[33][34][35][36]. Due to the promising results that were reported in these works, we used a Multi-layer perceptron (MLP) as described in [27,35].…”

Section: Neuronal Network γ-Ray Impact Position Predictionmentioning

confidence: 99%

Fast circuit topology for spatial signal distribution analysis

Lerche

Bosch

Spaggiari

et al. 2010

2010 17th IEEE-NPSS Real Time Conference

Self Cite

View full text Add to dashboard Cite

We present a novel active and analog readout and preprocessing topology for position sensitive photodetectors (PSPD) that allows to readout a large variety of PSPD devices with different pixel numbers. Additionally, the topology was designed to allow for a significant reduction of analog-to-digital conversion channels. The circuit topology replaces the common passive charge divider and consists of N input stages, N x M weighting stages and M analog adder stages, where N is the number of the input channels, i.e. the number of photodetector pixels and M is the number of outputs. The circuit performs the multiplication of a matrix (the weights) with a vector (signals). For this, the input stage makes M copies of each of the N input signals, the weighting stage multiplies these signal copies with N x M different weights and the output stage adds all weighted copies with the same copy index. For high flexibility, the weights are programmable and the topology allows to interconnect several identical circuits for larger N. As application for the circuit, we present a Neural Network based positioning scheme for γ-ray imaging detectors with thick, monolithic scintillation crystals. We used Monte Carlo simulation for training and evaluation of the method and found that the spatial resolution of the γ-ray imaging detectors was significantly enhanced. The strong border artifacts of the center of gravity positioning scheme in monolithic scintillation crystals could be corrected to a large extent.

show abstract

Section: Neuronal Network γ-Ray Impact Position Predictionmentioning

confidence: 99%

Fast circuit topology for spatial signal distribution analysis

Lerche

Bosch

Spaggiari

et al. 2010

2010 17th IEEE-NPSS Real Time Conference

Self Cite

View full text Add to dashboard Cite

show abstract

“…true interaction position and resulting outputs). Neural networks have been evaluated as either position estimators or direct LOR estimators by several groups [232][233][234] including EDNA [235][236][237] in the last two decades, since the popularization of FPGA technology that allows their real-time implementation. They achieve better resolution than Anger logic using the same set of 4 values and real-time hardware realizations are easy and inexpensive [235,238].…”

Section: Alternative Estimation Methodsmentioning

confidence: 99%

Development of a data acquisition architecture with distributed synchronization for a Positron Emission Tomography system with integrated front-end.

Varea¹,

Ramón²

View full text Add to dashboard Cite

Positron Emission Tomography (PET) is a non-invasive nuclear medical imaging modality that makes it possible to observe the distribution of metabolic substances within a patient's body after marking them with radioactive isotopes and arranging an annular scanner around him in order to detect their decays. The main applications of this technique are the detection and tracing of tumors in cancer patients and metabolic studies with small animals.The Electronic Design for Nuclear Applications (EDNA) research group within the Instituto de Instrumentación para Imagen Molecular (I3M) has been involved in the study of high performance PET systems and maintains a small experimental setup with two detector modules. This thesis is framed within the necessity of developing a new data acquisition system (DAQ) for the aforementioned setup that corrects the drawbacks of the existing one. The main objective is to define a DAQ architecture that is completely scalable, modular, and guarantees the mobility and the possibility of reusing its components, so that it admits any extension of modification of the setup and it is possible to export it directly to the configurations used by other groups or experiments. At the same time, this architecture should be compatible with the best possible resolutions attainable at the present instead of imposing artificial limits on system performance. In particular, the new DAQ system should outperform the previous one.As a first step, a general study of DAQ architectures is carried out in the context of experimental setups for PET and other high energy physics applications. On one hand, the conclusion is reached that the desired specifications require early digitization of detector signals, exclusively digital communication between modules, and the absence of a centralized trigger. On the other hand, the necessity of a very precise distributed synchronization scheme between modules becomes apparent, with errors in the order of 100 ps, and operating directly over the data links. A study of the existing methods reveals their severe limitations in terms of achievable precision. A theoretical analysis of the situation is carried out with the goal of overcoming them, and a new synchronization algorithm is proposed that is able to reach the desired resolution while getting rid of the restrictions on clock iii alignment that are imposed by virtually all usual schemes. Since the measurement of clock phase difference plays a crucial role in the proposed algorithm, extensions to the existing methods are defined and analyzed that improve them significantly. The proposed scheme for synchronism is validated using commercial evaluation boards.Taking the proposed synchronization method as a starting point, a DAQ architecture for PET is defined that is composed of two types of module (acquisition and concentration) whose replication makes it possible to arrange a hierarchic system of arbitrary size, and circuit boards are designed and commissioned that implement a realization of the architecture for the...

show abstract

“…Entre las aplicaciones de DPCs para read-out han aparecido muchas contribuciones destacables en la literatura reciente [12,191,221,222,261,279]. También se han propuesto estructuras de multiplexación paralela frente a la multiplexación en serie de las DPC, llamadas redes de división de carga simétricas (SCD, Symmetric Charge Division circuit) [261].…”

Section: Redes De Posicionamiento Discretizadounclassified

“…Una vez entrenada la red, se puede evaluar el error cometido sobre otro conjunto distinto (conjunto de test). Este es el método que se utilizó en esta Tesis y en diversas publicaciones previas [12,221,222]. En ellas se emplean redes que hacen uso de las señales procedentes del fotodetector (que pueden ser previamente preprocesadas para reducir su número y así simplificar la arquitectura de las redes) y la salida obtenida corresponde a las posiciones de incidencia estimadas.…”

Section: Red Neuronalunclassified

“…• Tanto en esta Tesis como en un cierto número de publicaciones ( [11,12,221,222]), se observa que las muestras que corrompen la predicción, posiblemente eventos desviados de otras posiciones que centellean en posiciones erróneas del cristal, pueden eliminarse mediante filtrado por energía. En este caso particular, en los histogramas 2D obtenidos tras el filtrado por energía aparecen "colas" alrededor de los máximos.…”

Section: Conclusiónunclassified

See 1 more Smart Citation

Redes neuronales y preprocesado de variables para modelos y sensores en bioingeniería

Mateo¹

View full text Add to dashboard Cite

High-precision position estimation in PET using artificial neural networks

Cited by 12 publications

References 12 publications

Fast circuit topology for spatial signal distribution analysis

Fast circuit topology for spatial signal distribution analysis

Development of a data acquisition architecture with distributed synchronization for a Positron Emission Tomography system with integrated front-end.

Redes neuronales y preprocesado de variables para modelos y sensores en bioingeniería

Contact Info

Product

Resources

About