l Medical data has four key components: pixel depth, photometric interpretation, metadata, and pixel data .
image size = header size ( including metadata ) + row x column x pixel depth x number of frames
l There are six main formats in radiological biological images: DICOM (Digital Imaging and Communication in Medicine), NIFTI ( Neuro Imaging Information Technology Initiative ) , PAR/REC (Philips MRI Scanning Format), ANALYZE (Mayo Medical Imaging ) , NRRD ( nearly raw raster data ) , in MINC format.
（1） DICOM –digital imaging andcommunications in medicine
Includes file formats and protocols for communicating using TCP/IP between entities capable of receiving images and patient data in DICOM format
A DICOM file contains the file header and *.dcm image data with the same file name. The size of the file header depends on how much information it provides. The file header contains: the patient's ID, the patient's name, the modality of the image, and other information, which defines the number of frames and the accuracy of the image.
（2） NIFTI—Neuroimaing InformaticsTechnology Initiative
The NIFTI format was invented for neuroimaging
The main feature is that it contains two affine coordinates that relate each voxel's index ( i , j, k) to its spatial position (x , y , z) .
The image raw data of NIFTI is stored as a 3D image
（3） NRRD—Nearly Raw Raster Data
The flexible NRRD format consists of a single header file and several image files that can be combined or separated. NRRD header files accurately represent N -dimensional raster information for scientific visualization and image processing .
The NRRD format characterizes Diffusion Weighted Image (DWI) volume and Diffusion Tensor Image (DTI) .
NRRD DWI and NRRD DTI data can be read as 3D slices in order to visually confirm that the orientation of the tensors is consistent with the expected neuroanatomical structure .
Ø Magic
NRRD0001: Initial and most basic version
NRRD 0002: Add key/value pairs ( key-value pairs )
NRRD 0003: Add " kinds ": field
NRRD 0004 : 加入“thicknesses”和“sample units”，added "thicknesses:" and "sample units" fields,general space and orientation information ("space:", "spacedimension:", "space directions:", "space origin:", and"space units:" fields) , and the ability for the "datafile:" field to identify multiple data files.
NRRD 0005: 加入“measurement frame”
Ø The symbol " \r\n " at the end of the sentence means the file created by Windows " \n " means the file created by the user;
Ø "< field >: < desc >" specifies information about a field, each line is called " field specification " or simply " field "
Ø "< key >:=< value >" key-value pair
The length of Key must be more than one character, value can be empty
ØComment lines start with #
The NRRD file header includes magic , field information, key-value pairs and comments. The file header is followed by a blank line, splitting the file header
Ø dimension: < int > dimension information representation
The number of samples on each axis is required information for each axis
sizes: <size[0]> <size[1]> ...<size[dim-1]>
size[i] represents the number of samples on the corresponding axis i , the axis is sorted from fastest to slowest (from fastest to slowest)
Ø type: <type>
type tags include int , float , unchar , block
Ø encoding: <encoding>
Describe the format of data writing, ascii , raw , hex
Ø Space：<spacce>
specify the specific space in which the array is located,
Space dimension: <int>
Space origin : <vector>
The center position of the first sample in the array, which is given first in the data file, or occupies the lowest memory address
Space directions: <vector[0]><vector[1]> <vector[2]>……<vector[dim-1]>
give the direction and spacing of each axis
（4） PAR/REC—Philips MRI scannerformats
（5） MINC—Medical Imaging NetCDF
MINC stands for Medical Image NetCDF Toolkit
MINC2 has been transformed from NetCDF to HDF5. HDF5 supports unlimited data types and is designed for flexible and efficient I/O and large-capacity complex data, which helps MINC2 process a large number of complex data sets.
