|
Preliminary clinical studies show that spiral CT scanning of the lungs can improve early detection of lung cancer in high-risk individuals. However, more clinical data are needed before public health recommendations can be made for population-based screening. Image processing algorithms have the potential to assist in lesion detection on spiral CT studies, and to assess the stability or change in lesion size on serial CT studies. The use of such computer-assisted algorithms could significantly enhance the sensitivity and specificity of spiral CT lung screening, as well as lower costs by reducing physician time needed for interpretation.
The intent of the Lung Imaging Database Consortium (LIDC) initiative is to support a consortium of institutions to develop consensus guidelines for a spiral CT lung image resource and to construct a database of spiral CT lung images. The investigators funded under this initiative are creating a set of guidelines and metrics for database use and developing a database as a test-bed and showcase for those methods. The database will be available to researchers and users through the Internet and will have wide utility as a research, teaching, and training resource.
Specifically, the LIDC initiative aims to provide:
- a reference database for the relative evaluation of image processing or CAD algorithms and
- a flexible query system that will provide investigators the opportunity to evaluate a wide range of technical parameters and de-identified clinical information within this database that may be important for research applications.
It is anticipated that this resource will stimulate further database development for image processing and CAD evaluation for applications that include cancer screening, diagnosis, and image guided intervention, and treatment. Therefore, the NCI encourages investigator-initiated grant applications that will utilize the database in their research. NCI also encourages investigator-initiated grant applications that provide tools or methodology that may improve or complement the mission of the LIDC.
See the Program Announcement: RFA: CA-01-001 LUNG IMAGE DATABASE RESOURCE FOR IMAGING RESEARCH 1 for more information.
Contact regarding programmatic issues:
Barbara Y. Croft, Ph.D.
Cancer Imaging Program, NCI
6130 Executive Blvd., Suite 6000
Bethesda, MD 20892
Telephone: 301-496-9531
Fax: 301-480-3507
E-mail: bc129b@nih.gov 2
Steering Committee
The consortium is guided by a steering committee consisting of 2 members from each of the five awarded institutions and 2 members from the Cancer Imaging Program.
LIDC Institutions
Cornell University 3
David Yankelevitz
dyankele@pop.med.cornell.edu 4
University of California, Los Angeles 5
Mike McNitt-Gray
mmcnittgray@mednet.ucla.edu 6
University of Chicago 7
Sam Armato
s-armato@uchicago.edu 8
University of Iowa 9
Geoffrey McLennan
Geoffrey-McLennan@Uiowa.edu 10
University of Michigan 11
Chuck Meyer
cmeyer@umich.edu 12
Mission
The mission of the Lung Image Database Consortium (LIDC) is sharing of lung images, especially low-dose helical CT scans of adults screened for lung cancer, and related technical and clinical data for development and testing of computer-aided cancer screening and diagnosis technology.
Principal Goals
To establish standard formats and processes for managing lung images and related technical and clinical data for use in the development and testing of computer-aided diagnostic algorithms.
To develop an image database as a web-accessible international research resource for the development, training, and evaluation of computer-assisted diagnostic (CAD) methods for lung cancer detection and diagnosis using helical computed tomography (CT).
| 1. |
The primary purpose of this project is to develop an image database for the evaluation of CAD methods for lung nodule detection and diagnosis using helical CT.
|
|
|
| |
1.1 |
The database is to contain helical CT images of representative cases selected
from lung cancer screening studies or diagnostic studies. The database should
enable the correlation of performance of CAD methods for detection and
classification of lung nodules with spatial, temporal and pathological ground
truth. The database is to be web-accessible by the imaging research community
as soon as possible. It should provide a resource for the training and
development of CAD methods. The Consortium will document evaluation metrics
that are valid for various CAD tasks as reported in the literature and that can
be used to assess investigator-developed CAD methods for lung nodule detection
and classification (benign/malignant). These quantitative methods are intended
to facilitate comparison of the relative performance of published CAD methods.
|
| |
1.2 |
The database is envisioned to be a single repository through which
investigators can:
(a) define subsets of data for individual research purposes using a query system, and
(b) define consistent reference subsets of data to evaluate the relative performance of CAD methods using a specified or recommended query method.
|
| |
1.3 |
The fields in the relational database should have sufficient detail to allow a
wide range of search parameters. All patient-identifying information will be
removed and the data anonymized, so that such information cannot be tracked.
Database fields should include, for example:
(a) specifications of the CT system used to generate the image and its image acquisition protocols;
(b) case parameters and reconstruction methods for representative normal and cancer
cases, including serial studies that are important for evaluation of CAD methods;
(c) spatial and pathological ground truth data, to allow a cross-correlation with computed results;
(d) the possibility for use of a flexible query system to allow for the evaluation of other future performance parameters for CAD, other image processing methods and related observer studies.
|
| 2. |
Secondary goals for this database include:
|
| |
2.1 |
Storage of raw CT image data, to permit exploration of alternative image
reconstruction methods or different reconstructed slice thicknesses that may
affect the performance of CAD methods, or to permit the evaluation of CAD
methods that incorporate the physical performance characteristics of the CT
system,
|
| |
2.2 |
Storage of images acquired through other tomographic modalities such as PET, to
explore improved means for classification of lung nodules, and
|
| |
2.3 |
Storage of digital chest images to permit the development and evaluation of CAD
methods for this modality.
|
Below are the LIDC committees and corresponding members.
Spiral CT Scanning, Dose, and Quality Control Committee
Mike McNitt-Gray (Chair)
Sam Armato
Heber MacMahon
Ella Kazerooni
David Yankelevitz
Geoff McLennan
Inclusion/Exclusion Criteria Committee
Sam Armato (Chair)
Heber MacMahon
Ella Kazerooni
David Gur
Mike McNitt-Gray
Geoff McLennan
Spatial Ground Truth Issues Committee
Chuck Meyer (Chair)
David Gur
Eric Hoffman
Larry Clarke
Lori Dodd
Mike McNitt-Gray
Sam Armato
Tony Reeves
Pathologic Ground Truth Committee
Geoff McLennan (Chair)
Local pathologists
Database Committee
Chuck Meyer (Chair)
Carey Floyd
Denise Warzel
Matthew Brown
Roger Engelmann
Sam Armato
Common Metrics; Software Tools Committee
Sam Armato (Chair)
Mike McNitt-Gray
Tony Reeves
Past Funding Opportunities Committee
Eric Hoffman (Chair)
Tony Reeves
Internet Availability Committee
Eric Hoffman (Chair)
Tony Reeves
Chuck Meyer
Mike McNitt-Gray
Mike Vannier
Industry Liaison Committee
Claudia Henschke (Chair)
Deni Aberle
Larry Clarke
Eric Hoffman
Mike McNitt-Gray
Ella Kazerooni
FDA Liaison Committee
Nick Petrick
Bob Wagner
Geoff McLennan
ACRIN Liaison Committee
Carl Jaffe (Chair)
Deni Aberle
Geoff McLennan
Publications Committee
Geoff McLennan (Chair)
David Gur
Barbara Croft
Statistics Committee
Mike McNitt-Gray (Chair)
David Gur
Bob Wagner
Lori Dodd
Charlie Metz
Jim Sayre
Heang-Ping Chan
Sam Armato
Curt Langlotz
Geoff McLennan
NIH Group
Larry Clarke
Barbara Croft
Dan Sullivan
Carl Jaffe
Houston Baker
Terry Yoo
Abraham Levy
Reports & Presentations: Lung Imaging 13
Reports and presentations produced for or as a result of CIP-supported workshops and other activities.
Publications: Lung Imaging 14
CIP publications that appeared in peer-reviewed journals or meeting proceedings.
Development of a CAD Assessment of a PN Database
David Yankelevitz, M.D.
Cornell University
Grant Number: U01CA091100
Lung cancer is the leading cause of cancer death worldwide, both in men and
women, with an estimate of over 164,000 new cases and over 156,000 deaths in
2000 in the United States alone. A principal reason for this high mortality is
that lung cancer typically is first detected at an advanced stage where the
prospects for cure are quite low. However, in those cases where it is found in
an early stage, the prospects for cure are quite high. Recognition of these
facts is a primary driver behind the development of improved screening and
diagnostic tools. We propose to form a collaborative group of institutions to
develop a large, high-quality internet-accessible spiral computed tomography
(CT) image database of pulmonary nodules. This will serve as an important
resource for researchers interested in developing improved methods for early
detection and screening for lung cancer.
Specifically this proposal plans to 1) develop the criteria for inclusion of
nodules within the database, 2) develop ground truth or pathologic diagnosis of
each nodule, 3) populate the database with the appropriate nodule candidates as
described above, 4) develop common data elements (CDEs) to classify each case,
5) develop criteria for measuring performance standards of various CADs, and 6)
develop an overall management plan for the consortium. The database developed
in this consortium will be an important resource for research and teaching
purposes. It will represent a standard that can be used for testing new CAD
systems. With the rapid advances in computer science and engineering, a
high-quality database that is continually evolving will be an invaluable
resource.
The design of this research proposal is somewhat novel in that we aim not only
to collaborate with others on the design and content of the image database, but
intend also to attach demographic and pathologic data to each case so that a
broad community of research can be served. Our overall management plan seeks to
aggressively identify collaborative partners from a variety of sources
including similar or related industry, for example the Visible Human Project.
Working groups will include radiology, CAD development, and informatics and our
outreach efforts will include patient advocacy and early users of the database.
Lung Imaging Database Resource for Imaging Research
Mike McNitt-Gray, Ph.D.
University of California, Los Angeles
Grant Number: U01CA091103
The aim of this research is to create a database resource for images that will
be used in analyses related to the detection and characterization of lung
cancer using spiral CT. There has been significant interest in the last few
years in using spiral CT lung scanning for lung cancer screening of patients at
high risk. Early detection and intervention may significantly reduce the
mortality rates of lung cancer and improve patient prognoses. In addition,
there is significant interest in the characterization of solitary or small
multiple nodules detected using lung cancer screening and conventional thoracic
CT exams. This is because the presence of nodules within the lungs is not a
reliable indicator of cancer. In fact, 50-80 percent of nodules detected by
current methods are benign; this percentage may even climb as smaller nodules
are detected with very sensitive screening techniques under consideration.
Therefore, detection of suspicious objects in the lung parenchyma, while a very
necessary step, is not sufficient for patient management. Additional imaging or
processing of the CT images may provide information that is useful in
establishing the diagnosis of the individual patient and determining the next
step in patient management. However, research in this area has been limited by
the difficulties in collecting cases on which image processing algorithms may
be robustly developed and tested. This is because it is difficult to establish
diagnostic truth for such key elements as lesion location and lesion diagnosis.
The establishment of a lung imaging database creates a resource for the
development and evaluation of methods for detecting and characterizing lung
cancer. When made available to researchers all over the world, this resource
would significantly reduce development time because it would allow imaging
researchers to focus on the their areas of expertise without having to focus on
case collection, establishing diagnostic truth and all of the other
infrastructure issues that detract from development. This database would also
allow direct and objective comparisons of techniques because common metrics
would be applied to identical cases. This will allow the image processing field
to move forward and to move from design to clinical implementation much faster.
The specific aims to accomplish this are: SA-1 To develop the necessary
consensus and standards for an image database resource related to the
detection, characterization and evaluation of lung cancer using spiral CT
imaging. SA-2 To construct, populate and test the database of spiral CT lung
image data and ancillary data including the information necessary about
diagnostic truth for each case. SA-3 To provide a means for documentation and
distribution of this database to researchers through the internet.
Standard Database for CT Lung Images
Samuel Armato, Ph.D.
University of Chicago
Grant Number: U01CA091090
The broad, long-term objective of this research project is to create a publicly available standard database of spiral computed tomography (CT) lung images. This lung image database will become an essential resource for the development of computer-aided diagnostic (CAD) techniques designed to help radiologists identify lung cancer in CT scans. The need for a standard lung image database is based on two recent developments. The first is the advancement of multi-slice CT scanners, which acquire images of multiple anatomic sections during each gantry rotation. Consequently, these scanners may generate an extensive amount of image data. The second development is the growing awareness among the American public and clinicians of the potential benefits of lung cancer screening using a low-dose spiral CT protocol. These developments are expected to dramatically increase the burden on radiologists. Moreover, primary interpretation from softcopy display will become a practical necessity.
What emerges from this scenario is a requirement for automated image processing methods that provide radiologists with quantitative information about suspicious abnormalities in the CT image data. Radiologists will then incorporate this information into their diagnostic decision-making process, with the expectation that cancer-detection sensitivity may be improved while decreasing both observer variability and interpretation time. Creation of a standard lung image database is critical to the endeavor of imaging research. This proposal addresses the important clinical and technical issues relevant to the creation of such a database.
The specific aims of the research are: (1) to identify the clinical requirements that must be imposed on a standard CT lung image database, (2) to address the technical issues and criteria involved with case selection for the CT lung image database, (3) to collect cases for the CT lung image database as a member of the Lung Image Database Consortium, (4) to develop strategies for the assessment of image processing and CAD methods using the CT lung image database, and (5) to investigate the effect of image reconstruction, multi-modality image registration, and registration of images acquired at different times on the utility of the CT lung image database.
As a member of the Consortium, we would demonstrate the flexibility necessary to reach consensus on the creation of a database that will serve as a standard resource for imaging research. The ideas presented in this proposal are expected to stimulate the efforts of the Consortium toward that goal.
Lung Image Databases with Pathologic Correlates
Geoffrey McLennan, M.D.
University of Iowa
Grant Number: U01CA091085
This application is in response to a specific request to establish a generalized CT-derived database representing ground truth in lung cancer and is not hypothesis-driven. Our broad goal is to help in the building of this database, and through that effort assist with the methodical development of appropriate lung cancer screening tools and protocols. Our group, with recognized experience in cooperative national projects, and with a broad perspective, will provide for the consortium: a well characterized group of study subjects with lung cancer, and with common lung cancer mimics such as histoplasmosis, supported by excellent radiologists and pathologists; expertise in the development of CT imaging protocols; a functional electronic transfer system for CT data sets from multiple sites, analysis and archiving of such data sets; expertise in DICOM standards, and in the issuing of web-based reports; methods for temporal matching of CT data points, important in the longitudinal follow-up of patients, and in matching excised inflated lobe data and histopathological data to the original patient CT; expertise in computational morphology, (i.e., the mathematical description of complex structures, their visualization, and their derived CT images).
We intend to apply this to a subset of resected lung tumors to help define pathological and CT ground truth. Image reconstruction algorithms. This is critically important for the identification and implementation of needed improvements in CT methods to maximize the chance of detection of subtle early lesions within the lung parenchyma and airways. Data from two different CT manufacturers multi-slice helical CT scanners. With mathematically derived virtual lung models, including early lung cancer development, for use in design of scanning and reconstruction methods.
Lung Image Database
Charles R. Meyer, Ph.D.
University of Michigan
Grant Number: U01CA091099
This grant application was awarded in response to RFA CA-01-001, the Lung Image Database Consortium (LIDC) Resource for Imaging Research. As a member of the LIDC will participate in formulating the multi-institutional lung imaging database acquisition and quality control specification, and begin collecting cases and populating a local database according to specifications resulting from the multi-institutional development of consensus guidelines.
Our clinical collaborators at the University of Michigan participating in this project have already had significant experience recruiting lung patients for another lung database project, the National Emphysema Treatment Trials (NETT). In this project Michigan ranked second in the number of patients screened for the study, and first in the enrollment of patients that passed the screen.
The Department of Radiology and the University Hospitals are committed to the acquisition of new generation CT and PET scanners over the duration of the LICD project. In direct support of the goals of a previously funded P01 as well as those of the LIDC, we have purchased a 4-CPU PowerEdge Dell server running RedHat Linux, configured with over 0.4 TB of RAID storage, all running on an uninterruptible power supply. The Raid and system disks are Past Funding Opportunities onto a LTO tape via ARCserveIT backup software. We have tested the installation of the Apache web server, PHP scripting language, and MySQL database. The system also supports the execution of AVS5, an application development environment for the manipulation and visualization of 3D data.
For the LIDC project we will implement the scrubbing of DICOM headers of unique patient identifiers, the population of a SQL database, and storage of associated CT datasets. Appropriate security and encryption has been addressed as well. The LIDC database will be available for public sharing through direct Internet access from our lab. The user web-interface will support identification of subsets of CT scans using SQL that may be downloaded for training/testing of CAD algorithms.
|
