You are on page 1of 1

Term Paper Assignment

Warmup exercise:
The pdf manual and Python scripts in this folder (“Term paper”) are for the term paper
assignment of PHY4283, Physics in Medicine. Follow the instructions in the manual to learn
about Generative Adversarial Networks and how they can be applied to virtual staining. Note
that this first portion will not be graded, but going through the exercise will be very helpful for
the graded portion below.

Assignment:
You will be graded on your ability to quantitatively assess the accuracy of virtual staining. You
will quantitatively compare virtually stained images with their physically stained counterparts
using metrics such as structural similarity index measure (SSIM) and mean square error (MSE).
Images are provided in the folder “high_resolution_images”. Note that in the exercise above you
were given code to compute peak signal-to-noise ratio (PSNR). Though PSNR is not a metric
that compares images, the Python code provided to you is very similar to that you can hopefully
write to compute other metrics such as SSIM. You may work with classmates and consult any
material (eg. lecture notes, websites), but you must submit original work. If you have questions,
you may ask the TA (Ms. Mengwan LI) or the course instructor (Dr. Condon LAU). Make sure
to submit the following via Canvas by 11:59 pm on Sunday December 4:
1. Scripts for computing PSNR and other metrics (eg. SSIM, MSE). You may use other
languages to do this assignment, but Python is recommended. (30 marks)
2. PSNR, SSIM, and etc. values for each pair of images. Note that an image pair is denoted
“virtual_” and “physical_”, followed by an alpha-numeric identification code. (30 marks)
3. PPT slides describing the algorithms for computing PSNR, SSIM, and etc., along with
numerical results and discussion items (eg. interesting observations from doing this
assignment). (30 marks)
4. A <200 words summary of your work, containing brief introductions, methodology, results,
and discussion. (10 marks)

Grading notes:
 For 1 and 2 above, if you only submit PSNR related material, you will not earn more than
50% on these two portions. More marks will be earned for computing more metrics (eg.
SSIM, MSE), but accuracy is important.

You might also like