Skip to main content Skip to secondary navigation

Image Fusion - Subtraction

Main content start

Developer: Vignav Ramesh

  • Algorithm & details

    Image fusion refers to the process of gathering all the important information from multiple images into a single, more informative image that is most conducive to human and machine comprehension.

    Given two DICOM image series of different modalities, the image fusion plugin fuses the corresponding slices in each series to create a final series containing the important features from each slice of the input series. The fusion of a sample CT slice and the corresponding PET slice is shown below:

    To perform image fusion, our algorithm uses the Principal Component Analysis (PCA) technique. The goal of the PCA algorithm is, when reducing from n dimensions to k dimensions, to find k vectors u(1), u(2), …, u(k) ∈R^n that span a linear subspace onto which to project the data so as to minimize the projection error. This dimensionality reduction aids in image compression.

    Using the PCA technique, our algorithm finds the eigenvalues of the image matrices and then fuses certain regions of the image together to maximize the points that are focused, resulting in higher eigenvalues. Similar to the discrete wavelet transform (DWT) technique, this fusion method satisfies some criteria of the eigenvalues—here, our criterion is the maximization of eigenvalues—of a specific region by choosing appropriate pixels from both images. The pixels from each image corresponding to the maximal eigenvectors make up the final fused image. The algorithm is shown below:

    After sorting the slices in each DICOM series by the z-value of their ImagePositionPatient attribute, each slice of one modality is fused with the corresponding slice in the other, and the resulting fused slices are combined in the output DICOM series.

    This plugin also performs image subtraction, that is, subtracting the pixel values of one image from the other for each pair of slices in the two input series. Image subtraction is used to identify key discrepancies between images. The image subtraction algorithm is shown below:

  • git Information

  • docker Information
    You can pull the image with : docker pull rvignav/docker-fusion