HI 5323 Session 11

Overview:

  1. Lecture: Medical Imaging (PDF file)
  2. Student Seminar: Image Processing in Medical Research (PDF file)
  3. Homework Assignment 4

Homework Assignment 4


Due at beginning of Session 13.

Download the paper Pablo Chacon and Willy Wriggers. Multi-Resolution Contour-Based Fitting of Macromolecular Structures. J. Mol. Biol., 2002, Vol 317, pp. 375-384.

(Assignment)

(Part A) Read the paper (the 'Materials and Methods' and 'Discussion' sections are particularly important) and note principles covered in the image processing course. This work is dealing with 3D matching but many concepts can be readily transferred from the 2D image processing case. Then write a summary adressing the following:

(Part B) In what contexts are the following concepts from the course used here: Fourier transform, convolution, FFT-acceleration, template convolution, Laplacian filter.

(Part C) What is different about 3D matching (more complex compared  to the 2D matching of images)? Hint: Meant is not the added dimension of the translational search, rather additional novel complexities in the paper that were not discussed in class.

(Part D) In the above paper the edge-detection properties of the Laplacian filter are used. What is the effect of the Laplacian on the performance of template convolution for matching of low-resolution 3D electron microscopy data (D1)? Application of the Laplacian operator in direct space is identical to multiplying in Fourier space with what function (D2)? Hint: Read the 'Discussion' thoroughly and read section 2.1.4 of this very nice filtering overview (if the link is broken a local copy is stored here). Based on this insight, is the Laplacian filter high-pass, low-pass, or band-pass (D3)?  Finally, a frequently used noise-reduction technique in electron microscopy is thresholding: density levels below the molecular surface threshold are set to zero,. In (D3) you should have answered what kind of noise frequencies are amplified by Laplacian filtering. So is thresholding a good idea to reduce the effect of noise when a Laplacian filter is used, or are there any caveats (D4)? Hint: This is not explained in the paper, you need to think. Consider that the Laplacian computes the second derivative in direct space. If there are problems can you suggest remedies?