Week 1: Diving Into MRI

This week, I completed my HIPAA training and learned more about my project and the other research being done by the radiology department at Mayo Clinic. In addition to my primary mentor, Dr. Panda, I will also be working with Dr. Zhang to set up and install the GPUs and other hardware components, as well as various programs such as QREADS, CUDA, and TensorFlow for image extraction, analysis, and machine learning. On the hardware side, each computer has 2 GPUs, facilitating a parallel processing computational infrastructure that will be indispensable to our machine learning operations with TensorFlow. On the software and programming side, I will mostly be using Python (3.6), and more specifically, Anaconda, which is the leading open data science platform powered by Python.

Although we are still working out the details, the main theme of my project is to streamline the diagnostic and treatment process for prostate cancer. The final goal is to help radiologists quickly develop personalized treatment plans that will optimize patient care and minimize patient discomfort. As part of this goal, I can also examine different types of radiotherapy to determine which one is optimal. Again, there are many paths this research could take.

For now, however, I am in the process of compiling patient information, MRI scans, and data from over 500 prostate cancer patients. This is the starting point of my project because I need to develop a substantial data set for computational work down the road. Most of the data set will be pulled from the Mayo Clinic in Rochester through a proprietary Mayo Clinic program called QREADS. Dr. Panda sent me an Excel spreadsheet listing all the prostate cancer MRI cases that we are interested in, and although this is not a complete list, there are thus far 267 cases that I must compile into my data set.

The actual process of creating my data set is simple, but arduous. The data I'm primarily interested in, besides the MR images, is the radiologists report, which is essentially composed of the date of the MRI, the type of exam, the indications (such as elevated Prostate-Specific Antigen, PSA), the date of the original report, the comparison, the impression (such as the condition of the surrounding tissue and presence of polyps and benign prostatic hyperplasia), and details about the prostate. Basically, I have to go through the listed case files to create my data set by adding text (txt) files in a folder I created on the shared physics drive. To do this, I have to open QREADS and search for cases matching the clinical number (ClinicNum). Since there are multiple results for each clinical number, I also have to match the dates to find the correct case. Afterwards, I copy and paste the radiologist report from each case into an individual txt file named AssnNum.txt, where AssnNum stands for accession number, a unique identifier used to track each patient order. For example, if a given case has a ClinicNum of 12-345-678 and an AssnNum of 12345678-9, then the radiologist report for 12-345-678 will go into 12345678-9.txt. Although this is simple work, it will take a long time because I have to manually create each file. Automating the process would not work because of the lack of an API or convenient user interface for QREADS, and because of security restrictions. Patient data must be secure.

After creating my data set, I will need to classify the different cases into two categories: screening and staging. There are two types of MRI exams for prostate cancer that we are interested in: screening exams and staging exams. While the screening process detects the presence of prostate cancer, the staging of prostate cancer presents greater detail, such as the threat level (Gleason score), tumor progression, and optimal therapy strategy. Staging MRI exams are more complicated, and naturally, take a longer duration to complete: 60 minutes on average versus 30 minutes on average for screening exams. But how can we tell, from an image set, if the MRI exam was staging or screening? Besides reading the radiologist report, the easiest way to make the distinction visually is to search for an endorectal coil in the MR images.

Siemens Sentinelle Endorectal coil

In staging exams, an endorectal coil is used because the technicians must get as close to the prostate as possible, since it's buried relatively deeply within the body and difficult to access. Therefore, the images from staging exams will show the presence of an endorectal coil, while the images from the screening exams will lack an endorectal coil. An endorectal coil will appear simply as a large darkly-colored rod, most easily seen in the axial T2 image series (ex: Ax Obl T2). Axial is simply the anatomical position of the patient during the scan and T2 stands for T2-weighted MRI, in which fluids appear bright (as opposed to T1, where liquids appear dark). T2 is the most useful for our purposes because it is most sensitive to pathology and will allow us to clearly see inside the prostate gland and detect tumors easily.

But why do we want to classify the cases into screening and staging categories? Well, it turns out that a significant part of my project will involve figuring out ways to improve the screening exam in order to make it as close as possible to the staging exam. This is because the staging exams cause a great deal of suffering and discomfort to the patients. In addition to having to lie down in an MRI machine for 60 minutes, patients undergoing the staging process also have to deal with the invasiveness of the endorectal coil. Ideally, we should only need one MRI exam to get all the information about a patient's prostate cancer, but the screening exam is not yet advanced or reliable enough to be used for the staging of prostate cancer. There is also the question of whether an endorectal coil is really necessary in the staging of prostate cancer. How much does it actually improve the results? The radiology department's initiative is to streamline the diagnostic and treatment process, and improving the screening exam to the level of the staging exam would go a long way toward accomplishing this goal. I do not yet know the specifics of what parameters we will use to determine and measure whether the screening exam results are becoming closer to the staging exam results, and therefore, I will need to consult with Dr. Kawashima, the resident expert on prostate cancer protocol. Moreover, there is a multitude of other factors, including the specific type of prostate cancer, the regions affected, and other complications specific to the patient, that I will need to learn to identify from imaging biomarkers. But in a nutshell, classifying the MRI prostate cases into staging and screening will enable us to optimize the screening prostate protocol.

Besides spending time in the physics lab, I was also able to shadow some nice and helpful MRI techs! They explained more about MRI, the various techniques to change image resolution, and the difference between T1 and T2. I was also able to observe several MRI exams, in addition to images from past cases. I was also quizzed extensively on anatomy and got to walk into "Zone 4", which is what they call the room that houses the actual MRI machine. The MRI machine had 1.5 Tesla magnets, which are always on, so any stray pieces of metal could be deadly. They are also planning on installing 3 Tesla magnets, which should improve the image resolution.


  1. Wow, Anthony! Although compiling your data set sounds arduous, writing software to help streamline treatment plans sounds like an incredible opportunity. It must be fascinating to work with patients and observe MRI's at the Mayo Clinic. I won't lie. The pictures of the apparatus needed to complete the exam looks a little uncomfortable, but I suppose it's a necessity. Great pictures! They brought the human element into your research. Can't wait to see what you post next week.

  2. Yes, the images were very helpful. I was wondering if there were other doctors/practicioners who were focused on improving the experience of those undergoing the screening exam, and to what extent your findings might overlap with theirs.

  3. Hi Anthony! You said the data entrance was simple but taxing, but it's very exciting to think about what it will all lead to. At the end, you mentioned you were meeting with Dr. Kawashima and that you had shadowed some MRI techs. Will you be continuing this out-of-physics-lab shadowing throughout your research?

  4. This post was very interesting Anthony! The images you have provided made it easier to understand your research and how the particular machines work. I am excited to see what you bring next week!

  5. Hi Anthony,
    Wow your project is amazing, and it sounds like you're making so much progress. You were talking about how your research will help radiologists develop personalized treatment plans. How long do you think it will take for radiologists to develop these plans after the research has been conducted?

  6. Hi Anthony! I'm very excited to see all the progress you've already made! I will say the images did indeed clear some things up and the last picture helped overall with the understanding of how MRI's work. How are you making sure that you are keeping all the data organized and correct?

  7. Hey Anthony!
    Everything you have done so far seems to fall perfectly in alignment with your plans. From what I can tell, your methods of collecting and organizing your data are very tedious but also well organized. The pictures you added were very helpful in clarifying the things I was confused. I can't wait to see how your project progresses! Good luck!

  8. Hi Anthony.
    I think it is really fascinating that you are trying to get the screening exam as close as possible to the staging exam. It seems as if this will be a great relief on the patient's part to be able to avoid the painfulness of the staging exam. How long do you suppose it will take for you to modify the screening exam? Also, which exam indicates the type of cancer and affected regions? Thank you for the explanation and pictures. I cannot wait to hear more!
    Zafeerah Sheikh

  9. Hello Anthony! Are there really "techniques" to improve image quality of an MRI? I mean wouldn't that require changing the strength of the magnet? And what are some of these techniques? Anyways, I'm really excited for data processing part of your project

  10. Wow, Anthony! It sounds like your first week was really fun and interesting. Thank you for the thorough explanations of what you are doing with the data you have, and the pictures were very helpful! Your research seems very well structured and intense. Can't wait to hear more about your experiences at Mayo!


Post a Comment