Artificial neural networks for positioning of gamma interactions in monolithic PET detectors

Abstract

To detect gamma rays with good spatial, timing and energy resolution while maintaining high sensitivity we need accurate and efficient algorithms to estimate the first gamma interaction position from the measured light distribution. Furthermore, monolithic detectors are investigated as an alternative to pixelated detectors due to increased sensitivity, resolution and intrinsic DOI encoding. Monolithic detectors, however, are challenging because of complicated calibration setups and edge effects. In this work, we evaluate the use of neural networks to estimate the 3D first (Compton or photoelectric) interaction position. Using optical simulation data of a 50 × 50 × 16 mm3 LYSO crystal, performance is evaluated as a function of network complexity (two to five hidden layers with 64 to 1024 neurons) and amount of training data (1000−8000 training events per calibration position). We identify and address the potential pitfall of overfitting on the training grid through evaluation on intermediate positions that are not in the training set. Additionally, the performance of neural networks is directly compared with nearest neighbour positioning. Optimal performance was achieved with a network containing three hidden layers of 256 neurons trained on 1000 events/position. For more complex networks, the performance degrades at intermediate positions and overfitting starts to occur. A median 3D positioning error of 0.77 mm and a 2D FWHM of 0.46 mm is obtained. This is a 17% improvement in terms of FWHM compared to the nearest neighbour algorithm. Evaluation only on events that are not Compton scattered results in a 3D positioning error of 0.40 mm and 2D FWHM of 0.42 mm. This reveals that Compton scatter results in a considerable increase of 93% in positioning error. This study demonstrates that very good spatial resolutions can be achieved with neural networks, superior to nearest neighbour positioning. However, potential overfitting on the training grid should be carefully evaluated.

Hybrid total-body pet scanners—current status and future perspectives

“…It has been published that modern readout architectures have implemented this hardware coincidence unit entirely in software due to the constantly increasing performance of computers [36,37,62]. This allows a higher degree of freedom in processing the data, in particular to use varying acceptance angles or more advanced methods from statistics or machine learning [20,29,67].…”

Artificial Intelligence for Monte Carlo Simulation in Medical Physics

“…NNs have also been used to estimate the twodimensional interaction position in the monolithic scintillator crystal in PET imagers, or a three-dimensional position when the depth of interaction (DoI) is estimated as well. The investigated NNs have yielded results with better spatial resolution [37,122], higher uniformity across the crystal volume [98] or faster implementation [149] compared to other existing methods (e.g. maximum likelihood [112] or nearest neighbours [141], among many others).…”

High-resolution monolithic LYSO detector with 6-layer depth-of-interaction for clinical PET

“…The main factors that could cause the performance differences between the two positioning algorithms are discussed in this section. The fraction of Compton scattered events is ~60% and has a large influence on the overall positioning performance (Decuyper et al 2021). The neural network is trained on many individual events while for the MNN database we use the mean of many signals which acts like a filter for PSFs of scattered events.…”

“…Most commonly used positioning algorithms are statistical ones, such as maximum likelihood estimation (Pierce et al 2018, Ling et al 2007, España et al 2014), k-nearest neighbour (van Dam et al 2011, Borghi et al 2016a, and more recently other machine learning algorithms like gradient tree boosting (Müller et al 2018) and neural networks (Wang et al 2013, Bruyndonckx et al 2004, Iborra et al 2019, Decuyper et al 2021. Multiple detector designs have been evaluated with respect to spatial resolution.…”