Question 1

Imagine being the network and hardware manager of an NHS hospital, and one day an academic       researcher asks you for advice on implementing a new segmentation algorithm for three-                    dimensional CT image data on hundreds of different images. She informs you that this algorithm can use either linear or non-linear image filters. As she is currently working from home, she asks if she    can be granted network access to use patient data while testing the algorithm. In addition, she only  has access to a dated personal laptop (single-core CPU @ 2.2 GHz, 4 GB RAM, 128 GB HDD, no           dedicated GPU), and wonders what options she has to make this project computationally feasible.    Briefly describe (<300 words) what advice you would give this researcher, including justifications.      Please ensure the following headings are used in your answer to facilitate the marking process:

1. Regulatory considerations;

2. Database access considerations;

3. Hardware considerations;

4. Algorithmic considerations.

Note: up to 4 marks will be awarded for the clarity and conciseness of your answer.         [20 marks]

Question 2

Computing in Medicine

For this question, you should assume that all of the computational analysis is performed on a personal computer with the following specifications: 4.40 GHz 4-core CPU, 8 GB RAM, 500 GB SSD, 2 GB GPU. (1 GB= 10003  Byte, 1 MB= 10002  Byte)

a.           After running a simulation using MATLAB, the following error was received: “Out of

memory. Type HELP MEMORY for options.” MATLAB calculated that 13.8 GB of memory  required to run the simulation. List two potential solutions that would enable getting the results of this simulation. In addition, for each solution, name a factor that might              prevent you from running your simulation using that solution. The solutions could             involve use of the computer mentioned above, but additional resources can be                  considered.                                                                                                                                   [2]

b.           An MRI dataset is saved in the NIFTI file format.  The header is NOT a separate file, and   the size of the header is 348 bytes. The 3D scalar image volume comprises 256×256×191 voxels. The volume is stored using short (2-byte) integers. Calculate the uncompressed   file size, then calculate how many NIFTI files can be stored in each of the different types of memory in the computer with the specifications above. (You should consider both      volatile and non-volatile memory, and assume that all memory is available for storing     NIFTI files.)  Clearly show your working.                                                                                 [3]

c.           Assume your friend uses another personal computer with the following specifications:     4.40 GHz 1-core CPU, 8 GB RAM, 500 GB SSD, 2 GB GPU. When comparing the                    computation speeds of these computers, it was found that the first computer is faster,    but not four times as fast even though the first computer has a four-core processor. Give three reasons why the first computer is not four times faster than the second.            [3]

Using the digital Boolean circuit and the input binary signals a, b, c shown in Figure 1 below:

d.           Find the outputs u, v and the intermediate outputs x, y and z, for the following cases:

(i)          a = 0, b = 1 and c = 0; (ii)         a = 1, b = 1 and c = 0.

For each case, clearly show your working.                                                                             [2]

Find the Boolean algebra expressions for u and v as functions of arguments a, b and c.   [2]

Figure 1. A Boolean circuit.

What two steps are needed to calculate two’s complement of a binary number? In  addition, calculate the two's complement for each of the four following 5-bit binary

numbers, and clearly show your working:                                                                              [2]

01001                 01011                 00111                 00001.

Digital Signal and Image Processing

Figure 2 shows a 6-by-6 image I and two structuring elements K and H.

Figure 2. Simple image I and two structuring elements (K and H)

Calculate the magnitude and phase of the gradient at pixel (2,2) when a Sobel edge

detector filter is used.                                                                                                                [4]

Suppose a simple edge detector has been devised for image I based on thresholding.   Compute the output binary image when the threshold for edge detection is set to 125.

[1] Based on the image obtained in (h) and considering the two structuring elements K and

H shown in figure 1, calculate the following:                                                                         [7]

1.   I ● K

2.   (I ⊝ K) n (  ⊝ )

3.   (I ⊝ H) − (I ⊕ H)

What is the purpose of the morphological operation (I ⊝ K) n ( ⊝ )? What is the      name of this operation? Name one medical application where this operator can be used, and explain your answer.                                                                                                           [3]  For the image obtained in (h), calculate the distance transform using the city-block and   quasi-Euclidean methods, and comment on their advantages or disadvantages.          [3]

Question 3

Computing in Medicine

a.           The Chief Executive Officer of a small local hospital recently attended a course and has heard about EHR systems and PACS. The hospital mainly cares for patients with chronic long-term conditions. It does not have an emergency, radiology or cardiology                  department, but they plan to purchase an MRI system in the next few years. List the      advantages and disadvantages of implementing each system (two advantages and two disadvantages for the EHR system; one advantage and disadvantage for the PACS).        Ensure motivation is given for all advantages and disadvantages, and make sure the      advantages and disadvantages are evaluated in the context of this particular hospital.

[6]

Two tests A and B were applied to two groups of 100 people respectively, with both groups sampled from two populations which were differently affected by a disease.  The confusion matrix obtained   from both tests are shown below:

b.           Calculate prevalence of the disease for both tests and comment on the differences.  [2]

c.           Calculate the sensitivity and specificity, and comment on the differences and what they reveal about the tests A and B.                                                                                                 [2]

d.           Calculate the positive prediction values (PPV) and negative prediction values (NPV) for  both tests. Explain the general importance and meaning of these two statistics.          [2]

e.           Comment on the differences in PPV and NPV observed for both tests in d) and draw       conclusion as to the reason(s) for the differences in view of the results found in b) and  c).                                                                                                                                                    [1]

Digital Signal and Image Processing

An adaptive Weiner filter is a pixel-wise low pass filter that employs an estimate for the local mean  and variance to remove additive Gaussian noise. The mean (u) and variance (G 2 ) are calculated over a window of size N × M around a pixel in the i-th row and j-th column of the image I as:

N−1              M−1

Eq. (1)

and

i+   2           j+   2

G 2 (i, j) =      ∑           ∑      I2(n, m) − u(i, j)

N−1                M−1

The denoised image J is calculated on a pixel basis as:

J(i,j) =  u(i,j) +  (I(i,j) − u(i,j))

where v 2  is the noise variance.

Eq. (2)

Eq. (3)

f.            Figure 3 shows an image where the noise variance is equal to v 2  = 3580.

Figure 3. Image I

(i)         Calculate the grey-level value of filtered image J for the pixel located at   (i, j) = (1, 1). Assume that the size of the local window is 3 × 3 and the   image has been zero-padded.                                                                         [2]

(ii)        Repeat (i) assuming that the pixels outside the image have replicated the

values at the edges and corners.                                                                    [2]

g.          When the noise variance and the signal variance are comparatively similar, the Weiner filter is reduced to a 3 × 3 kernel defined as:

1       1       1

18    18    18

1       5       1

H =        

18      9      18

[        ]

(i)         Apply the filter H to image I assuming all the pixels outside the image      have the value zero.                                                                                          [2]

(ii)        Repeat (i) assuming that the pixels outside the image have replicated the

values at the edges and corners.                                                                    [2]

h.          Is the adaptive Weiner filter linear or nonlinear? And is the filter H defined in (g) linear or nonlinear? Indicate the reasons why the filters are linear or nonlinear,  and comment on the advantages of disadvantages of either case.                      [2]