Skip to content

Latest commit

 

History

History
73 lines (56 loc) · 1.61 KB

README.md

File metadata and controls

73 lines (56 loc) · 1.61 KB

Automatic Target Recognition (ATR) for CT-based Airport Screening System

Description: Project to classify CT images as target or non-target.

EECE 5890: Machine Learning for Image Processing

Professor: Dong Hye Ye, Ph.D.

Authors

Resources

Python

NiBabel - read / write access to some common neuroimaging file formats
PyNifti - older version of NiBabel

Visualization

MIPAV - quantitative analysis and visualization of medical images

Details

Targets

  • Saline
    • 3.5%, 10%, 15% concentrations
  • Modeling (polymer) clay
  • Rubber sheets: ¼” thickness (minimum) + other rubber in bags

Non-Targets

  • Food
  • Drinks
  • Electronics
  • Magazines
  • Containers not filled with saline

Dataset

Cropped CT image for each segmented object

Target labels

Non-target:0, Saline:1, Rubber:2, Clay:3

Image Quality

Many artifacts which lead to imprecise density, volume, mass, shape

Possible features

  • Mass
  • Mean
  • Standard deviation
  • Histograms
  • Higher-order moments
    • Skew, kurtosis, entropy
  • Texture
    • Wavelets

Possible classifiers

  • PCA
  • SVM
  • Decision Tree
  • Adaboost
  • Deep neural network

Performance metric

PD = # targets detected / # targets scanned
PFA = # false alarm objects / # non-targets scanned

Goal: PD > 90%, PFA < 10%

Data Files

Nifti file format: Standard Neuroimaging File Format
.nii.gz: gzipped image