Virtual Imaging Clinical Trial for Regulatory Evaluation
Note All code is under CC0 (Creative Commons 1.0 Universal License) except for the FBP code that is a derivation of a GPL3 code and remains as GPL3.
UPDATE Aug 17, 2022
- You can also access the VICTRE automatic pipeline that puts together all the steps on a single Python class.
- We created a sample VICTRE full pipeline app on precision FDA that you can access here (needs registration).
OPPORTUNITY ANNOUNCEMENT / We are seeking a scientist specialized in computational modeling and simulation of medical imaging systems. Successful applicants will have demonstrated expertise in modeling complex physics and biological systems in the areas of imaging including magnetic resonance and x-ray tomographic methods, and in the multi-scale in silico modeling of anatomy, physiology and pathology. The selected candidate will join an interdisciplinary research team seeking to demonstrate the benefits of in silico methods in the regulatory evaluation of imaging products (see, for instance, JAMA Netw Open. doi:10.1001/jamanetworkopen.2018.5474). For more information, please contact Aldo Badano (aldo.badano@fda.hhs.gov).
Clinical trials are expensive and delay the regulatory evaluation and early patient access to novel devices. In order to demonstrate an alternative approach, a recent effort at the Division of Imaging, Diagnostics, and Software Reliability at the U.S. Food and Drug Administration (known as the VICTRE project) demonstrated the replication of one such clinical trial using completely in-silico tools and compared results in terms of imaging modality performance between the human trial and the computational trial. The VICTRE trial involved imaging approzimately 3000 digital breast models in digital mammography and digital breast tomosynthesis system models. On this page we are making all the in silico components of VICTRE freely available to the community.
Citation: ''Evaluation of Digital Breast Tomosynthesis as Replacement of Full-Field Digital Mammography Using an In Silico Imaging Trial.'' Aldo Badano, Ph. D., Christian G. Graff, Ph. D., Andreu Badal, Ph. D., Diksha Sharma, M. Sc., Rongping Zeng, Ph. D., Frank W. Samuelson, Ph. D., Stephen Glick, Ph. D., and Kyle J. Myers, Ph. D. JAMA Network Open. 2018;1(7):e185474;doi:10.1001/jamanetworkopen.2018.5474.
Overview of the VICTRE project A 1-hour summary presentation of the project and findings was given at the FDA Grand Rounds on 3/14/2019 and can be found here.
VICTRE team: Aldo Badano, Ph. D., Christian G. Graff, Ph. D., Andreu Badal, Ph. D., Diksha Sharma, M. Sc., Rongping Zeng, Ph. D., Frank W. Samuelson, Ph. D., Stephen Glick, Ph. D., and Kyle J. Myers, Ph. D.
This software and documentation (the "Software") were developed at the Food and Drug Administration (FDA) by employees of the Federal Government in the course of their official duties. Pursuant to Title 17, Section 105 of the United States Code, this work is not subject to copyright protection and is in the public domain. Permission is hereby granted, free of charge, to any person obtaining a copy of the Software, to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, or sell copies of the Software or derivatives, and to permit persons to whom the Software is furnished to do so. FDA assumes no responsibility whatsoever for use by other parties of the Software, its source code, documentation or compiled executables, and makes no guarantees, expressed or implied, about its quality, reliability, or any other characteristic. Further, use of this code in no way implies endorsement by the FDA or confers any advantage in regulatory decisions. Although this software can be redistributed and/or modified freely, we ask that any derivative works bear some notice that they are derived from it, and any modified versions bear some notice that they have been modified.
VICTRE replicates various steps in silico of the comparative clinical trial. The pipeline consists of 9 components:
-
breastPhantom - anthropomorphic breast model
Breast models (in silico subjects) are generated using a procedural analytic model. The model allows for varying patient characteristics including breast shape, glandularity and density, and size. -
breastCompress - model of physical breast compression
Breast models are then compressed with use of open-source finite-element software FEBio (separate download). -
breastCrop - phantom cropping
Breast models are then cropped to a fixed volume to speed up loading and fit them in the limited Graphics Processing Units (GPU) memory. -
breastMass - cancer mass model
Breast masses are generated using a procedural generation code. This model allows for variations in mass size, shape, amount of spiculations, and other characteristics. -
Lesion insertion
Lesions are then inserted in a subset of the compressed breast phantom population to create cancer cases. The lesion insertion locations are randomly chosen from the list of possible locations given by the breast phantom generation code. The selected location is then passed through checks to ensure that the lesion is within the phantom boundaries, non-overlapping with tissues like air/muscle/nipple/skin, and non-overlapping with already inserted lesions. This code is available under Lesion Insertion. This code is developed using Python. -
X-ray imaging
These in-silico patients are then imaged using a state-of-the-art Monte Carlo x-ray transport code (MC-GPU). We obtained projection images for the two modalities in the VICTRE trial: full-field digital mammography (DM) and digital breast tomosynthesis (DBT). Details on downloading and running this module can be found at MC-GPU code. This code is developed using C and CUDA. -
Reconstruction
VICTRE implemented an filtered back-projection (FBP) reconstruction algorithm for DBT using single-threaded C based on Fessler's cone beam computed tomography (CBCT) reconstruction toolbox. We modified an extension of the single-threaded C code developed by Leeser et al. for reconstruction of CBCT projections to allow for DBT reconstruction. The modifications account for an x-ray source moving in an arc about the object with a stationary detector with the z-axis of the object normal to the detector plane. Code and instructions available under FBP DBT reconstruction in C. We also have the same code available in Matlab, available at reconstruction code using Matlab. -
Regions/Volumes of interest (ROI/VOIs) extraction
After the images are acquired, lesion-present and lesion-absent ROIs are extracted from the DM images (or VOIs from the DBT volumes). For extracting lesion-absent ROIs, we applied rigorous checks including if the subimages are within the reconstructed volume boundaries and non-overlapping, to find appropriate locations. This code is available under ROI Extraction. This code is developed using Python. -
Reader models
The ROIs are then interpreted by in silico readers using a location-known exactly paradigm. Code and instructions of use can be downloaded from reader models. This code is written in Matlab.
VICTRE image datasets in DICOM format are available for download at the Cancer Imaging Archive. The images include DM projections, DBT projections and reconstructed volumes. VICTRE images are converted from Raw to DICOM (Matlab code available at raw to DICOM conversion). Metadata such as the image generation are added as the DICOM tags to allow for reproducibility. DICOM tags include patient information, clinical trial description, imaging study performed per modality and series under each study, and breast type, lesion absence or presence, and compressed breast thickness are included as attributes of the patient.
Digital mammography projection ROIs dataset is now available at https://github.com/DIDSR/VICTRE_DM_ROIs.
Sample phantoms from the VICTRE trial are now available under Sample-phantom-data.
-
VICTRE configuration files and parameters
Contains the configurations files and command line parameters used in the VICTRE trial for generating and running the four types of breast phantom densities (dense, heterogeneously dense, scattered density and fatty) through the pipeline. This will be useful for anyone trying to replicate a part of or the entire VICTRE pipeline. -
Locations
Contains phantom lesion locations in voxels and corresponding DBT locations. The digital mammography ROI locations are available at https://github.com/DIDSR/VICTRE_DM_ROIs. -
Sample phantom data
Contains raw data for one phantom from each of the four breast density categories. -
Raw to DICOM conversion
Contains Matlab function to convert VICTRE raw data to DICOM.