联系我们： 手动添加方式: 微信>添加朋友>企业微信联系人>13262280223 或者 QQ: 1483266981

MPHY0020 / 2023-24 (Main) Page 1 TURN OVER
A6U
UNIVERSITY COLLEGE LONDON
EXAMINATION FOR INTERNAL STUDENTS
MODULE CODE : MPHY0020
ASSESSMENT Pattern: MPHY0020A6UE
MODULE NAME : MPHY0020 – Computing in Medicine
LEVEL: : Undergraduate
DATE: : 13-May-2024
TIME : 14:30
DURATION : 3 Hours
Late submission is permitted for Controlled Conditioned exams but late
penalties will apply – any submissions that are up to 40 minutes late will
be penalised, after which no submissions will be accepted under any
circumstances.
You must ensure to allow sufficient time to upload and hand in your work
This paper is suitable for candidates who attended classes for this module
in the following academic year(s):
Year
2022/23
Duration 3 Hours
Additional time for converting handwritten
notes to PDF where applicable 10 Minutes
Upload window 20 Minutes
Total time
3 Hours 30 Minutes
Additional material N/A
Special instructions N/A
TURN OVER
MPHY0020 / 2023-24 (Main) Page 2 CONTINUES
A6U
Exam paper for MPHY0020 Computing in Medicine
Answer all parts of all questions 1-6.
The number in square brackets on the right-hand side after each question indicates
the maximum number of marks allocated for that part of the question. You can use this
as an indication of the extent and depth of the answer expected.
Answers can be hand-written or typed using a word processor, or you can use a
combination of both methods.All answers must be submitted electronically via the
AssessmentUCL platform. Submissions via email or any other method cannot be
accepted.
Everything you submit, including drawings, diagrams, and mathematical
calculations, must be your own work and not copied or reproduced from any other
source.
You must not discuss examination questions or your answers with any other person,
including other UCL students, or obtain assistance from any third party during the
period of the examination.
MPHY0020 / 2023-24 (Main) Page 3 TURN OVER
A6U
QUESTION 1
You are performing segmentations on hundreds of medical images locally on a workstation,
where the image data is accessed through an XNAT server. Suddenly, your GPU breaks
down mid-computation and is fully unusable, but your operating system has handled this
failure and is still up and running, meaning your results computed up to the breakdown are
still in the memory.
(a) What steps would you need to take to ensure that the medical images are secure and
not at risk of corruption or loss, and why [2]
(b) How do you ensure that the segmentations completed prior to the hardware failure are
not lost and do not need to be repeated after the workstation is repaired [2]
It turns out that the GPU cannot be replaced as the motherboard of your workstation is so
old that compatible GPUs are no longer sold, yet you do not have sufficient budget to fully
replace the workstation.
(c) Given that these segmentations were the final computations required to complete your
thesis, what solution would you propose to enable completion of the final remaining
computations Explain your reasoning and list two issues you might encounter. [2]
QUESTION 2
A small hospital trust has recently implemented a PACS system, but due to budget
limitations is not able to provide a sufficient number of workstations to allow all clinicians to
access, view and share the PACS data. Instead, the hospital administrator realised that all
staff members carry their own mobile devices, and they are considering whether to have
staff use their own mobile devices to access the PACS data.
(a) In 100 words or less, discuss which data protection regulations apply (if any), and
whether these regulations would allow for this solution. Explain your reasoning. [4]
The hospital decides to proceed with this plan. One of the clinicians carries a smartphone
with 64 GB of free storage capacity, 4 GB of memory (combined RAM and GPU memory),
and a microSD Flash memory card with a 128 GB capacity.
(b) Based on these specifications, how many 3D MRI images can this clinician store on
their device in volatile memory, and how many in non-volatile memory Two answers
are expected; one for the number of images in volatile memory, one for the number of
images in non-volatile memory. Assume that each 3D MRI image contains
128x128x512 voxels, each voxel is represented by a uint16 integer, the images are
uncompressed, each image is stored in uncompressed NiFTI format, and 1 GB = 109
bytes. Show your working. [4]
MPHY0020 / 2023-24 (Main) Page 4 CONTINUES
A6U
QUESTION 3
Figure 3.1 shows an ECG signal measured by a medical device starting from time = 1.46
s until just before = 2.21 s. During this period, = 750 samples are recorded. It is known
that typical ECG frequencies fall in the range of 0.5-150Hz.
(a) What are the sampling frequency and sampling period of the medical device For
the ECG signal, explain whether (or not) aliasing occurs in the measurement. [3]
(b) Assuming Discrete Fourier Transformation (DFT) is performed on the recorded
discrete ECG signal, calculate the corresponding resolution in the frequency domain
( ). If one wishes to obtain a frequency resolution ≤ 0.5 Hz, compute the required
duration of the recording. Show your working. [3]
(c) In practice, other signals and noise also contribute to the measurements additively,
contaminating the ECG signal. Typical sources include high-frequency (≥600Hz) noise
commonly caused by thermal noise and muscle tremor, and signal drift caused by the
patient’s breathing. Assume the normal respiratory rate for an adult at rest is 12-18
breaths per minute.
i. Briefly (<100 words) describe the artifacts caused by these additional signals in the time domain, and indicate time points in Fig. 3.1 where these additional signals are evident. [2] ii. Assuming a sampling frequency = 2500 Hz, sketch the frequency spectrum of such an ECG signal that is compromised by those artifacts, indicating the location of each component and the cut-off frequency of the spectrum. [4] (d) Describe the process of filtering required to remove the irrelevant signal and noise components specified in (c) and recover the ECG signal. Make sure you include and explain the appropriate threshold frequencies. [4] 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 Time (s) -2 0 2 4 Fig. 3.1: a sampled ECG signal Amplitude (a.u.) MPHY0020 / 2023-24 (Main) Page 5 TURN OVER A6U QUESTION 4 A 60-year-old male submitted a blood sample for routine testing, and this sample tested positive for prostate cancer. Understandably worried, he asks his doctor what is the chance of him actually having prostate cancer, having received this positive test result. (a) The doctor explains that the test was performed on 1000 independent blood samples. Assuming a disease prevalence of 3% and a test sensitivity of 80%, how many samples corresponding to men who actually have prostate cancer correctly receive a “positive” test result In other words, how many True Positives (TP) occurred Show your working. [2] (b) Further assume that the test specificity is 94%. How many samples corresponding to healthy men nevertheless receive a “positive” test result In other words, how many False Positives (FP) occurred Show your working. [2] (c) State the confusion matrix for this test applied to these 1000 samples. [1] (d) Based on your previous answers, compute the chance that, after receiving a positive test result, a patient actually has prostate cancer. Show your working. [2] MPHY0020 / 2023-24 (Main) Page 6 CONTINUES A6U QUESTION 5 Consider the logic circuit shown in the following figure. (a) For the logic circuit depicted in Fig. 5.1, derive the truth table stating the outputs X and Y for all possible combinations of Boolean inputs A, B, and C. Show your working for at least one set of A/B/C values. [4] (b) Find the Boolean algebra expressions for X and Y as a function of arguments A, B and C. Show your working. [2] (c) Find outputs X and Y for the cases where bits A, B and C encode the following two integer numbers: +3 and -5. Show your working. [2] Figure 5.1: a Boolean logic circuit. MPHY0020 / 2023-24 (Main) Page 7 TURN OVER A6U QUESTION 6 In digital image processing, a mean filter can smooth images, restoring the intensity of each pixel by computing the average of the surrounding pixels and assigning the average to it. Given the 2D input image ( , ), pixel ( , ) of the filtered image is calculated within the filter kernel of size × as ( , ) = 1 ∑ ∑ ( , ) + 2 1 = 2 1 + 2 1 = 2 1 . (a) When filtering with a 5 × 5 filter kernel, write down the equation for calculating ( , ) and illustrate the kernel with its coefficients. [3] (b) Is the above filter kernel separable, and why Explain and quantify the advantage of using a 2D separable kernel. [4] (c) Compared with the local mean filter, the non-local means (NL-means) filter better preserves image details while denoising (smoothing) the image. NL-means filtering measures the mean of neighbouring pixels, weighted by pixel-wise similarity coefficients. The NL-means filtered image at pixel ( , ) can be calculated with an × kernel from the input image as ( , ) = ∑ ∑ ( , ) ( , ) + 2 1 = 2 1 , + 2 1 = 2 1 where the weight ( , ) measures how similar the surrounding pixels are to the target pixel ( , ), as given by ( , ) = 1 , | ( , ) ( , )| 2 2 . Here 2 is a filtering parameter and , is a normalisation constant ensuring that the weights within the kernel sum to unity, i.e., ∑ ∑ ( , ) + 2 1 = 2 1 + 2 1 = 2 1 = 1. i. Is the NL-means filter kernel linear or nonlinear, and why In addition, list two aspects in which linear filters are preferred over nonlinear filters, and justify your choices. [3] Eq. (6.1) Eq. (6.2) Eq. (6.3) Eq. (6.4) MPHY0020 / 2023-24 (Main) Page 8 END OF PAPER A6U QUESTION 6 - continued ii. Figure 6.3 shows an enlarged section (right) of a brain MRI (left). Assuming zero-padding of the image, calculate the signal intensity of NL-means filtered image for the pixel at the top-right corner ( , ) = (1,5), using a 3 × 3 NL means filter kernel with 2 = 10,000. [4.5] (d) Briefly describe and explain an advantage of using a Gaussian kernel instead of a finite difference approach in edge detection. Explain how edges may be found using the first and second derivatives of Gaussian filters. [1.5] Figure 6.3: Brain MRI image (left) and enlarged section (right).