E2E-differentiable Charged Particle Tracking Data
DOI10.5281/zenodo.12759188Zenodo12759188MaRDI QIDQ6722445FDOQ6722445
Dataset published at Zenodo repository.
Ralf Keidel, On Behalf of the Bergen Pct Collaboration, Nicolas R. Gauger, Tobias Kortus
Publication date: 18 July 2024
Copyright license: Creative Commons Attribution 4.0 International
In this data repository, we extend an earlier dataset (https://zenodo.org/records/7426388) generated for the Bergen pCT DTC by various additional Monte Carlo (MC) simulations (generated using the Gate 9.2 simulation toolkit [1, 2] built upon Geant4 [3,4,5]) with different setups and phantom materials with particular focus on an E2E-differentiable tracking algorithm provided on GitHub. The data repository is further extended by trained model checkpoints and results of addtional analysis required for running the code, without re-training and re-evaluation. Training Files Files: We provide multiple simulations for different phantom geometries and simulation setups, each generated with a mono-energetic pencil beam (230 MeV, 2 sigma). The supplied files include single beam spots for water phantoms of various thicknesses (100, 150 and 200 mm): water_{100,150,200}_5k.npz (validation data) water_{100,150,200}_10k.npz (test data, taken from https://zenodo.org/records/7426388) water_{100,150,200}_100k.npz (training data) Columns: All above-mentioned simulation files contain MC simulated data of a single simulation run in tabular form, where each row represents a single particle hit inside the detector. Furthermore, each particle hit is parametrized by the following columns: posX, posY, posZ: Measured x, y, z position (in millimeter) of the particle hit relative to the simulation origin defined by the center of the phantom. edep: Amount of energy (in MeV) deposited by a particle while interacting with the sensitive area of the detector. eventID: Each primary is simulated in its own isolated "event" and gets an incremental ID. Everything that happens during the simulation of said primary is grouped under the same eventID. Events are simulated independent of each other. trackIDs are only unique within their respective event. trackID: A track describes a single particle throughout its entire lifetime in the simulation. In any given event, the first track (trackID = 1) is always associated with the primary particle. Every subsequently produced secondary particle has an incremental trackID. parentID:The parentID specifies the trackID in the current event that caused this track to exist. If the parentID is 0, the particle is a primary, i.e., generated by the particle beam. Otherwise, the row describes a secondary which was generated through interactions of a primary with the traversed matter. volumeID[2]: Incremental numerical identifier of layer containing particle hit inside GATE volume 2 defined within the detector geometry. 0 for tracking layers, 1 for calorimeter layers. volumeID[3]: Incremental numerical identifier of layer containing particle hit inside GATE volume 3 defined within the detector geometry. Unique identifiers (starting from zero) for tracking layer (0, 1) and calorimeter layer (0, 1, , 40). Model Checkpoints and Analysis Data Files: Additionaly to the simulated track data, we provide the trained model checkpoints and additional analysis data, generated using the source code for end-to-end differentiable charged particle tracking published on GitHub. The compressed directory contains the following directories: cka: Calculated CKA similarities [6] estimated for all trained model combinations . mode: Minimum energy connecting curves (mode connectivity) [7] of Bezier splines optimized for all trained model combinations. pat_lambda_{25,50,75}: All trained end-to-end differentiable tracking networks. ptt: All trained two-step tracking networks. [1]S. Jan, G. Santin, D. Strul et al., GATE -Geant4 Application for Tomographic Emission: a simulation toolkit for PET and SPECT,Phys Med Biol. Phys Med Biol, vol. 49, no. 19, pp. 45434561, 2004. [2]S. Jan, D. Benoit, E. Becheva et al., GATE V6: A major enhancement of the GATE simulation platform enabling modelling of CT and radiotherapy, Physics in Medicine and Biology, vol. 56, no. 4,pp. 881901, 2011. [3]S. Agostinelli, J. Allison, K. Amako et al., GEANT4 - A simulation toolkit, Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, vol. 506, no. 3, pp. 250303, 2003. [4]J. Allison, K. Amako, J. Apostolakis et al., Geant4 developments and applications, IEEE Transactions on Nuclear Science, vol. 53, no. 1, pp. 270278, 2006. [5]J. Allison, K. Amako, J. Apostolakis et al., Recent developments in geant4,Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, vol. 835, pp. 186225, 201 [6] Kornblith, S., Norouzi, M., Lee, H., Hinton, G. (2019). Similarity of Neural Network Representations Revisited. 36th International Conference on Machine Learning, pp 61566175, 2019. [7] Garipov, T., Izmailov, P., Podoprikhin, D., Vetrov, D., Wilson, A. G. (2018). Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs. Advances in Neural Information Processing Systems, pp. 87898798, 2018.
This page was built for dataset: E2E-differentiable Charged Particle Tracking Data
