bergtholdt/spie16lung

Pulmonary Nodule Detection Using a Cascaded SVM Classifier

This repository contains the data used for producing the results in the SPIE16 paper " used in the SPIE16 Paper" (9785-038).

There are two alternative version. In volumes_json you can find all Volumes as JavaScript Object Notation (JSON) encoded deep nested data structure including Cliques and Annotations. These contain a lot of fields not necessarily needed by 3rd parties. A subset of relevant fields and individual tables can be found in the *.csv files (comma separated values) or alternatively in the *.json files. There is one for table for each: Volumes, Cliques, Annotations and Contours. Note that some fields are encoded as lists of strings. If you load the csv-file in Excel, some rows containing long strings are not read correctly.

• volumes.csv / volumes.json

Field	Description
PatientID	the unique LIDC/IDRI PatientID
Dataset	"Training", "Verify" or "Test" (Verify is also used for Training)
SeriesUID	DICOM tag SeriesInstanceUID
StudyUID	DICOM tag StudyInstanceUID
ImageType	list of DICOM ImageType tags
FrameOfReferenceUID	DICOM tag FrameOfReferenceUID
XRayTubeCurrent	DICOM tag XRayTubeCurrent
KVP	DICOM tag KVP
ImageFileNames	list of original dicom file-names used in this volume (in order)
ImageUIDs	list of dicom SOPInstanceUID of each slice

• annotations.csv / annotations.json

Field	Description
AnnotationID	the unique (internal) AnnotationID
Primitive	0 for Nodules and NonNodules, 5 for NoduleVolume
AnnotationData	either bounding box for NoduleVolume or position and radius for Nodule/ NonNodule (DICOM Patient Coordiantes)
VolumeID	the unique VolumeID
HUMax	maximum Hounsfield unit in Annotation
HUMean	mean Hounsfield unit in Annotation
HUMin	minimal Hounsfield unit in Annotation
HUStd	standard deviation of Hounsfield units in Annotation
Label	a label prefixed by the type NoduleVolume, Nodule or NonNodule, followed by the original label from LIDC/IDRI XML data
AnnotationsIncluded	for NoduleVolume, AnnotationIDs of contours that are included in this NoduleVolume
AnnotationsExcluded	for NoduleVolume, AnnotationIDs of contours that are excluded in this NoduleVolume

• contours.csv / contours.json

Field	Description
AnnotationID	the unique (internal) AnnotationID
Primitive	2 for contours (input to NoduleVolumes)
AnnotationData	the triplets of contour coordinates in DICOM patient coordiantes
VolumeID	the unique VolumeID
ImageUID	dicom SOPInstanceUID of the corresponding slice
SlicePosition	the slice number within the volume (starting from 0)
Label	a label prefixed by the type NoduleContour, followed by the original label from LIDC/IDRI XML data

• cliques.csv / cliques.json

Field	Description
CliqueID	the unique CliqueID
PatientID	the unique LIDC/IDRI PatientID
VolumeID	the unique VolumeID (SeriesInstanceUID plus a number to force uniqueness within dicom series)
AnnoCount	number of Annotations in the Clique
NoduleCount	number of Annotations labeled as "Nodule < 3mm" in this Clique
NoduleVolumeCount	number of Annotations labeled as "Nodule >= 3mm" in this Clique
NonNoduleCount	number of Annotions labeled as "NonNodule" in this Clique
Annotations	list of AnnotationIDs as cross reqference to Annotation table
DiameterMean	mean diameter of annotations in this clique
Diameters	list of diameters
DiceMean	mean of all pairwise dice coefficients
Dices	all pairwise dice coefficients
HUMax	maximum Hounsfield unit in Clique
HUMean	mean Hounsfield unit in Clique
HUMin	minimal Hounsfield unit in Clique
HUStd	standard deviation of Hounsfield units in Clique
VolumeMean	mean of all Annotation's volumes in $mm^3$
Volumes	the Annotation's volumes in $mm^3$