AI System Documentation                          Introduction

This site provides general information on the AI research servers for CT image analysis.

The main documents are accessed by clicking on the links in the sidebar to the left.

Sidebar Documents

Name	Description
Conventions	Basic conventions for labeling and image indexing
Medical Images	The three image formats for 3D CT images: DICOM, NifTi and V4 and converting between them
Image Access	Curated image datasets are best accessed by the vdimage command which translates between image-codes. DICOM series-ids, and file-system image locations.
Image Analysis	The system standard image analysis algorithms that can be applied to curated image datasets
Nodule Analysis	Analysis programs for pulmonary nodules
Docker Access	Information on how to run image analysis algorithms on the system that involve dockers.
Utility Programs	Python utility programs that are available to programmers. These are mainly chatbot generated and may be used as system commands (without the .py) or installed directly in users’ programs. They are in available in ~reeves/v4/py/.
Servers	Relevant Server details
Dataset Curation	Image datasets are curated to make all dataset images conveniently available. A description of the curation process.
Vsimba	A tutorial on how to use the on-line web-based image viewer vsimba.
Curated Results	A table showing the current status of our research analysis with results available for different datasets

 

Location

The CT image datasets and v4 software is installed on all research servers.

The remainder of this introduction provides general background information on the system

 

Accessing Research dataset images

Image Referencing

A unique naming system is used for every image (CT scan) in the research datasets.  While the DICOM unique series identifier (DSID) is often used for this, the system used here is much shorter, directly provides dataset and case identification and deals with image datasets in which the DSID is not available.

Image Reference Format the image-code

The syntax for an image-code that identifies an image within a curated image dataset  is:

 <dataset-id><case>-<image>

Where:

<dataset-id> is a one- or two-character identifier for for the research dataset

<case> is the dataset case identifier

<image> is a unique identifier for each image within a case

For example, the code “NL00023-s1” refers to the dataset NL (NLST_Test) dataset case 000023

 and image labeled s1 within that case. 

Accessing Dataset Image data

A system command vdimage is used for image access. The basic function of vdimage is to determine the full image path name when provided with an image-code. For data set images that have DICOM unique series identifiers, vdimage will do a reverse look-up and return the image code associated with that series id.

Curated Research Datasets

The following image datasets are currently available on the research servers

Dataset Code	Cases	Images	Dataset Contents
W	50	50	The VIA-ELCAP public nodule dataset. cases have all lung nodules documented
NT	2493	6715	The NLST_Test dataset. This contains images selected from the NLST dataset to test the Sybil cancer prediction model
UN	3405	40698	Pamplona image dataset up to 2015. Also, recent Pamplona
ME	6672		All of the MESA CT images, CAC for series 1-6 and Lung for series 5
DL	1614	1614	Duke Lung Dataset