10. Basic 3D DICOM Dataset EDA

Heading

DICOM Volume Dataset EDA

ND320 C3 L2 08 Volume Dataset EDA

Summary

Methods for dataset analysis basically boil down to using the same tricks as you’d do for individual volume analysis and being on the lookout for inconsistencies in data.

Inconsistencies usually boil down to two classes:

  • Clinical anomalies - the things related to either anatomical anomalies like missing organs, pathologies like tumors or implants such as limb prosthesis, ports/cannulas, surgical implants, presence of contrast media, etc. Sometimes these things can result in artifacts in the images, so it’s good to be aware of them

  • Informatics anomalies - things related to specifics of data acquisition or variations in DICOM encoding coming from different scanners. These would be things like slice spacing consistency, image dimensions, variations in photometric encoding, etc

Basic knowledge of DICOM and intuition for what things could go wrong are always useful when analyzing the datasets. I will post some examples of great dataset EDA at the end of this lesson as well.

Check for Understanding

What would be relevant things to look at when analyzing a medical volume dataset?
SOLUTION:
  • Pixel spacing tag
  • File names
  • Image Orientation Patient and Image Position Patient tags
  • Photometric Interpretation tag
  • Series Description tag
  • Rows/Columns tags

WS Intro

Below you can find the Jupyter Notebook that has been used in this walkthrough.

Code

If you need a code on the https://github.com/udacity.

Next Concept