How to read DICOM A simple question? •How do radiologists do their work? •How does “data” factor into their work? Radiologist analyze data Report Image data PDF Radiologist How data factor in? • Like any other discipline, they must make valuable insights by analyzing data. Data Insight Too much data, too little time • 9-10 hour shifts: • 100s of cases per day • Sample workload: • 70 x-ray scans • 10 CT scans • 2 MRIs • Trauma cases: • 3,500 images to 4,500 images • Lung CT: • 200-400 images Upstream Data Preprocessing • Loading Data • DICOM (.dcm) • Extracting Pixel Values • Nifity (.nii) • Plot values as image or • HL7 graph What is DICOM? • Data structure and network protocol • File extension • .dcm • .dicom • Stores image data • Interoperability -> Transmits, stores, retrieves, prints, processes, displays medical images, network protocol = Digital Imaging and Communications in Medicine Data Structure • Key – value pair • (Group, data element) • (0028, 0011) -> Columns • (0028, 1052) -> Rescale Intercept • Image , Rescale Intercept http://dicomlookup.com/getpdf2016.asp DICOM (tag group, element group) 0002, 0010) Transfer Syntax UID UI: Explicit VR Little Endian (0002, 0012) Implementation Class UID UI: 1.3.6.1.4.1.9590.100.1.3.100.9.4 (0002, 0013) Implementation Version Name SH: 'MATLAB IPT 9.4' ------------------------------------------------- (0008, 0005) Specific Character Set CS: 'ISO_IR 100' (0008, 0008) Image Type CS: ['ORIGINAL', 'PRIMARY', 'AXIAL', 'HELICAL'] (0008, 0012) Instance Creation Date DA: '0' (0008, 0013) Instance Creation Time TM: '174459.843' (0008, 0016) SOP Class UID UI: CT Image Storage • https://www.leadtools.com/help/sdk/v20/dicom/api/overview-basic-dicom-file-structure.html • http://dicom.nema.org/dicom/2013/output/chtml/part05/sect_3.10.html Preamble 128 byte preamble characters DICM Prefix File Meta Information Data element What is DICOM? • Digital Imaging and Communications in Medicine • Image in DICOM format • .dicom or .dcm • 2D grayscale image • Transmits, stores, retrieves, prints, processes, displays medical images, network protocol • used by all forms of imaging: X-ray, CT, MRI, Ultrasound, PET scans, etc. CT image is composed of cross sectional slices are stocked 200 to 400 slices CT image is composed of cross sectional slices are stocked 200 to 400 slices What is Pixel data & Metadata? • Each slice has it’s own metadata For Example: • A data stream of the pixel samples that comprise the Image • Pixel data describes a color image with a single sample per pixel (single image plane) • Take the pixels into NumPy array • Pydicom: package for DICOM metadata editing • Metadata will tell us slice dimensions and CT scan dimensions Meta data Validate it Pixel size Y • field of view (the size in centimeters of the area presented by an image) • Pixel size : (the size in Pixel size X millimeters of the x and y coordinates of a pixel) • Slice thickness : (the size in Slice thickness Z millimeters of the z coordinate of a pixel) How does “data” factor into their work? Window – Height & Width Greyscale value range 0 - 255 More bits larger grey scale range Hounsfield Units = HU “In a CT scan, Hounsfield Unit is proportional to the degree of x-ray attenuation and it is allocated to each pixel to show the image that represents the density of the tissue.” 100 kW – 80 kW = 20 kW Attenuation Co-efficient is 20 kW Hounsfield units are used to describe X-ray the absorption rate of X-rays that was 100 kW 80 kW measured for each voxel when the CT Detector scan was taken. Pixel size : Hounsfield units: The size of absorption rate Attenuation Co-efficient is 20 kW the pixel is in of X-rays millimeters. 0.05 mm Linear attenuation Coefficient “Characterizes how easily a volume of material can be penetrated by beam of light, sound, particles, or other energy or matter.” “attenuated” means (weakened) so the beam was weaken as it pass through material. E.I. energy was lost. HU value for air is -1000 HU HU value for water is 0 HU HU value for blood is 35-40 HU HU is a linear transformation of original linear attenuation coefficient measurement Slope Intercept Y=mx+b Slope-intercept form If X = 77.137497 m = 1.211 b = -64.434 Our HU value is 28.979 Normalization of the pixel data HUs conversion Patient Comparing data for diseases Patient 1 Patient 2 Patient 3 Why? Factors of variations “many of the factors of variation influence every single piece of data we are able to observe” –Deep Learning Book • Type of CT scanners • The size of the patient • Voltage on the X-ray tube GE Health CT scanner Different Hardware Different Gray Scale Values output Philips CT scanner normalization of ratings means adjusting values measured on different scales to a notionally Normalize common scale. output By HU machine learning inference Siemens CT scanner statistical comparison Deep learning models output DICOM Caveats Encoding • The stored pixel_array (voxel array) of each slice in a CT volume are the Hounsfield values of this slice. • Hounsfield units originally were intended to be used with a 0 - 4,000 range (hence some older machines, use 12 bit pixels to store pixel_arrays --> 12 bit = 0 - 4,095). • Nowadays almost all DICOM sub formats use 16 bits for the pixel_array LOOK UP THE MANUFACTUR Homework – watch this YouTube video! • https://www.youtube.com/watch?v=KZld- 5W99cI&ab_channel=PetraLewis&t=s By Guido Zuidhof • http://199.116.233.101/index.php?title=Main_Page • https://ai.stanford.edu/~syyeung/cvweb/tutorial1.html
Enter the password to open this PDF file:
-
-
-
-
-
-
-
-
-
-
-
-