In order to understand and predict the biological effects caused by ionizing radiation, many studies based on Monte Carlo (MC) simulations have been
dedicated to simulate radiation-induced DNA damage. The stochastic nature of radiation makes MC calculations a well-suited method for simulating energy
deposition in tissues, and subsequently an effective method to evaluate and predict the biological effects induced by ionizing radiation. As a generic
open source Monte Carlo toolkit, Geant4 and Geant4-DNA extension provide an open-source platform that allows users to simulate biological damage induced
by ionizing radiation at the DNA scale. In our work, a single cell model with detailed DNA geometry was used to predict the yield of DSBs produced by
ionizing radiations considering the contributions from direct and indirect effects from the physical, physicochemical and chemical stages of radiation
interactions in cell.
Different organization levels in the model of DNA geometry in our single cell model.
The geometrical model of cell nucleus and DNA in the single cell model was developed based on the wholeNuclearDNA example in Geant4. As shown in above
figure. an ellipsoid of half-axes 11.82x8.52x3.00 micrometer was determined for the nucleus model, which was then filled with a detailed DNA geometry to represent
the complete human genome.Except the elaborate and explicit model of DNA geometry to quantify direct DNA damage from the energy deposition in physical
stage, another important characteristic in this single cell model was the DNA molecular chemical model using the G4Molecule class of Geant4-DNA to record
the early indirect DNA damage which is generated in a few nanoseconds after physical stage. In chemical stage, only the reactions between hydroxyl radicals
and 2-deoxyribose were considered to be the candidates of indirect damage on DNA, as shown in the following table.
Related reactions of the Geant4-DNA chemical module used in the chemical stage simulation.
Different irradiation sources were simulated in Geant4 Monte Carlo codes to generate physical and chemical interaction points in single cell model.
For example, mono-energetic protons from 2 MeV to 100 MeV. As shown in the following figure, the yield of DSB per Gy per Gbp induced by mono-energetic
protons with different energies in our single cell model were in general agreement with other published validated results.
The simulated DSB yields by mono-energetic protons with increasing kinetic energy in our single cell model, as well as the results published
in some similar studies
This work has been published this year, you can find more details in the following paper:
Gold nanoparticles (GNPs) have attracted interest in the past few years for their use as CT contrast agents (CA) due to their high X-ray attenuation, simple
surface chemistry, and biocompatibility. Dose enhancement due to the low energy secondary electrons generated by interactions with GNPs was also observed
in dose measurements during CA imaging studies. Since then, multiple efforts of dose enhancement through the use of GNPs have been put forth.However, the
observed dose and DNA damage enhancement from GNPs in experimental studies have significantly differed from most computational predictions. The understanding
of the radiosensitization mechanism of metallic nanomaterials and the accurate prediction of this type of biological effects, remains a big challenge in
radiobiology research. To investigate the radiosensitization properties of GNPs and better understand the intricate DNA damage induction mechanisms involved
in GNP-aided radiotherapy, different kinds of GNP models were applied in our work.
Simulation geometry of the computational cell model in Geant4.
In 2019, a simple Geant4 single-cell model was built considering a realistic cell and organelle geometry in our lab, as shown in the above figure.
For GNP geometry, gold shperes were used to represent GNPs, and they were randomly distributing in a virtual shell with 1 micrometer thickness and abutting
to the surface of nucleus or cellular wall, as well as in the cytoplasm. In this work, the radiation transport simulation code for cell irradiation is
developed based on the microdosimetry example in Geant4 and a mixed-physics approach is taken using Geant4-DNA physics to model the low-energy interactions
within the cell, and Livermore physics to model the interactions in the GNPs. The energy deposition points inside nucleus were selected as DNA strand breaks
according to randomly sampling model. Then, possible double strand breaks (DSBs) were determined according to the DBSCAN algorithm.
Dose enhancement for three different GNP distributions under: (a) 100 keV mono-energetic photon irradiation, (b) 6-MV FFF photon irradaition.
However, the absence of chemical stage simulation and indirect damage calculation limited this study about the GNPs-mediated enhancement in SSBs and
DSBs yield. Therefore, we developed a new single cell model with complex multi-scale DNA geometries and chemical process simulation to simulate direct
and indirect DNA damages from ionizing radiation. You can find more details about this single cell model in my website. After the single cell model was
built considering a realistic cell and DNA geometry in nucleus, gold spheres were used to represent the GNPs with different size in the cell model. Different
parameters of GNPs, particularly the size, quantity, and location, were investigated in this work. A analysis process was applied by using DBSCAN clustering
algorithm to calculate the number of DSBs from the two kinds of strand breaks (SBs): direct and indirect SBs.
a) Simulation geometry of the single cell model with 12000 90 nm GNPs in cytoplasm. b) 12000 GNPs congregated around the nucleus. c) 8207 GNPs were simulated in nucleus.
In this work, the number of DSB per cell with same simulation history (10 billion) were computed to study the different GNPs radiosensitization in
different scenarios. The contributions of direct and indirect effects in DNA damage were investigated by the three categories of DSBs: (a) Direct DSBs are composed
by direct SBs only, (b) Indirect DSBs are composed by indirect SBs only, (c) Mixed DSBs are composed of direct SBs and indirect SBs.
Left plot: DSB enhancement factor with different GNPs locations in 100 keV mono-energetic photon source, and the contributions of direct/indirect effect. Right plot: Enhancement of DSB, direct SB, and indirect SB with three different GNP size in 100 keV photon source.
One paper of our GNPs work has been published in 2019, another one has been submitted:
Xiandong Zhao, Ruirui Liu, Tianyu Zhao, Francisco J. Reynoso. "Modeling gold nanoparticle radiosensitization from direct and indirect action in a
complete human genome single cell model in Geant4." (submitted).
Date:
September 2020
Data Analysis in Proton Therapy
Artificial Intelligence was widely used in several field in past three years, radiology is no exception. Last year, a 3D dose-prediction model for Photon therapy was developed by Shiraishi and Moore. They uses artificial neural networks to predict the dose distribution around tumor tissue. Compare to Photon therapy, Proton therapy's features are totaly different. However, we can also use neural networks to do the dose-prediction when we approve enough knowledge and information to this machine learning model.
Schematic plot of a sample artificial neural networks.
The original datasets are stored in CT image form. In these CT images, we can get the planned dose distributions for 5 prostate and 10 brain proton-therapy cases. To get the input data for my neural networks, I need to get the dose value, relative position to health organs and PTV (planning target volume) for each voxel. Here, I used Pydicom and Scikit-Learn to do the preprocessing of CT images. For example, get the contour data for each organs and PTV.
The top two plots show the contours of organs and CTV/PTV in prostate and brain cases. The bottom table shows the parameters (input data) of my neural networks for prostate cases.
The neural networks model is built by TensorFlow. After I got all voxels' information from all the given cases, these datasets were randomly allocated, with three-fourths used for training, and one-fourth used for validation. The following plots show the training results for both prostate cases and brain cases. As you can see, the accuracy (dose error ratio) is very tiny in high dose area of prostate cases, and the dose error reaches to lower than 20% in all dose range of brain cases.
Training results for prostate casesTraining results for brain cases
Date:
April 2018
Proton Therapy Simulation
Most people may not be familiar with particle physics (another name is high energy physics), especially unfamiliar with particle accelerator, which is a most frequently used machine in particle physics research. However, nearly everybody know proton therapy, which is an example of particle accelerator's application. In this project, I will use Monte Carlo Simulation in Geant4, to simulate the proton therapy.
Hadron Therapy simulation example in Geant4. In the right side, there is a box that can be filled with water to simulate human tissue. A small detector can get different information (ex. the dose deposited in voxels) in specific position.
The reason why the proton beam can be used as surgical knife to destroy tumor, is the specific attributes of proton's ionizing radiation energy loss during its travel through matter. There is very few dose delivered to tissue during the proton's travel, util the last few millimeters before proton is stopped (loss its all kinematic energy). Based on this phenomenon, medical personnel use the particle accelerator to target a tumor with different depth by a beam of proton with different initial energy. The following figure shows the Geant4 simulation of a single proton's energy loss during its travel.
This figure shows a single proton's energy deposited with different depth in water (human tissue). This proton (initial kinematic energy is 100 MeV) is stopped at about 75 millimeter finaly. The narrow peak known as a Bragg peak.
After the simulation for single proton, I used Geant4 to simulation thousands protons to get the energy deposit distribution in a human tissue. The accelerators in proton therapy typically produce protons with energies in the range of 70 to 250 MeV. In Geant4, we can set the value of initial kinematic energy of proton, to control the energy of proton beam easily. The nozzle size of proton therapy determine the sigma (width) of proton beam. We can also set it easily in Geant4.
This figure shows three thousands protons interacting with water (human tissue), the kinematic energy of the proton beam is 70 MeV, sigma of the proton beam is 8 millimeter. The blue line is the trajectory of proton. The green line is the trajectory of gamma, which is generated by proton's inelastic scattering. The red line is the trajectory of electron, which is generated by gamma's compton scattering.
According to the Monte Carlo simulation, I got the energy deposit distribution in the tissue slice (2 millimeter thickness). Once we get that distribution in the Monte Carlo model to match the real measurement of proton therapy, we can import patient CT images into this Monte Carlo model for the dose calculations.
This figure shows the energy deposit distribution in water (human tissue) on the longitudinal cross section. The kinematic energy of proton beam is 226 MeV, and the sigma of proton beam is 8 millimeter. There are 100k protons in this simulation.
Date:
October 2017
Magnetic Focuser
As a powerful mutant, Magneto can manipulate magnetic field to achieve a wide range of effects. We know magnetic field can change the momentum and track of charged particles. So, if Magneto was real, the researchers at CERN can save a lot of money on Large Hadron Collider (LHC). In reality, people spent millions of dollars to build a powerful Quadrupole Magnets system, to focus the proton beams in LHC.
Super Proton Synchrotron is used to focus the proton beam, and the protons for LHC are accelerated in it. The left top plot is 3D model of the complex quadrupole magnets design. The right bottom plot is the schematic diagram about how the system appears to work: The black arrows show the direction of the magnetic field, the red arrows indicate the direction of the force on a proton, and the yellow arrow shows the direction of proton.
To focus the high energy protons at tera electronvolts (TeV) level, the LHC need to use a powerful quadrupole magnet with superconducting technology. The magnetic field arround the superconducting coils can reach 12 tesla. This is amazing! We know the Rare-earth Magnets are the strongest permanent magnets so far, but only can reach 1.4 tesla. However, if we just want to focus low energy charged particle beam, why not give a chance to rare-earth magnets? In this project, we need to focus the alpha particle with 5.488 MeV kinematic energy and therefore the rare-earth magnet is expected to go far.
The left bttom plot shows the quadrupole magnetic field which is simulated by ANSYS. The left top plot shows the CAD design of the frame to hold the permanent magnets. The right plot shows the frame made by CNC.
In this project, I simulated the magnetic field by ANSYS, design a frame to hold the permanent magnets by PTC Creo, and simulated the whole focusing process of alpha particle by Geant4. You can check the Geant4 simulation of focusing alpha beam in the bottom plot. In nature, the radiation of alpha source is isotropic. According to the simulation in my project, the sensor can receive more signal (increase 134%) after deploy the permanent magnets system!
The Geant4 simulation: we can import the ANSYS magnetic field simulation result in Geant4, and then simulate the movement trajectory of alpha particle.
Date:
May 2017
MTx & MTRx Module
With the development of Computer Aided Design (CAD) and 3D printing, more and more people prefer choosing the products based on personalization and customization. People in scientific research field are not immune as well, especially when the needs of research are becoming increasingly different from industry. There is a dichotomy between the academic world and the industrial world. Therefore, CAD and 3D printing technology are now widely used in research. This project is an example.
This plot shows the modules used in industry: LC connector (left top), the transceiver module for TOSA and LC Connector (left bottom). Both of them exceed 6 mm in height.
When we try to use the LC connector and related transceiver module to connect TOSA, we find their heights are exceed our requirement, 6 mm. We cannot find any module meet our needs on the market. Therefore, I need to develop the mid-board miniature dual channel optical transmitter (MTx) and transceiver (MTRx) by myself. After several revisions, I designed MTx and MTRx modules, and they will be used for detector front-end readout of the ATLAS Liquid Argon Calorimeter (LAr) trigger upgrade. For more detail, please see my paper: "Mid-board miniature dual channel optical transmitter MTx and transceiver MTRx". The following figure shows the CAD plots of MTx and MTRx.
The design of the latch and the coupling fiber to the TOSA.
These mid-board optical modules can also be used in other systems where mid-board optical coupling is needed. Both MTx and MTRx are ready for production by injection molding. This work is supported by the US-ATLAS Research and Design program for the upgrade of LHC, and CDRD grant from U.S. Department of Energy.
