Hunting for more Lesions!

I got back to work this week. My mentors were still away for Spring Break, so I mostly worked alone. Continuing the work of week 4, I spent my time reviewing patient MRI scans. As I mentioned in a previous blog post, my task for each MRI scan is to search for legions indicative of Hepatocellular Carcinoma (HCC). I then mark these lesions as a region of interest and compare it to a separate region of interest located in normal liver tissue, which acts as a control. 

I got the first few patients done pretty quickly. It was very easy to identify tumors. But I soon discovered that these patients were the exception, not the standard. For most of the patient scans, the visual difference between lesions and surrounding, healthy tissue was very subtle. Gradually, I trained my eyes to notice these slight differences in color and shading, but the entire process was based on trial and failure nevertheless. Additionally, some of the scans were extremely low-quality and pixelated, making lesion detection even more difficult. Other scans were corrupted and had missing data, so I was forced to work around that as well.

Another difficulty lay in the variety of naming and operational conventions used to produce the patient MRI scans. My goal is to examine several types of scans: Apparent Diffusion Coefficient (ADC), Diffusion Weighted Imaging (DWI), Axial In-phase, Axial Out-phase, Axial Pre/Post-phase LAVA, and Axial T2. Each one reveals a different perspective. However, the exact names differ with each radiologist, so it took some effort to match everything together. Also, some patients lack certain scans, either because the radiologist in charge of the exam did not believe them to be necessary, or were limited by their equipment. 

However, despite these setbacks, I'm making steady progress. I expect to finish lesion identification for all patients by the end of next week, which will allow me to move on to the next phase of my project. More will be revealed in next week's blog post!

Comments

  1. Hi, Richard. You have told me that you would like to discover a way to remain youthful eternally. Has your research made your more or less optimistic on your prospects of achieving that goal?

    ReplyDelete
  2. This comment has been removed by the author.

    ReplyDelete
  3. Thanks, Richard. I hope you had a lot fun for spring break. How did you work around the missing/corrupted data scans? It seems impossible (since they're missing/corrupted). Did you have software or some super secret technique you can't share?

    ReplyDelete
  4. Hi Richard! Thank you for explaining all the medical scan lingo. Where do you stand in reaching your final goal? What are you most looking forward to discovering in the upcoming weeks? Looking forward to more!

    ReplyDelete
  5. This comment has been removed by the author.

    ReplyDelete
  6. Hello Richard! I am wondering if there is another type of scanning that is easier to use and does have little error? Thank you and i have enjoyed your posts!

    ReplyDelete
  7. Hi Richard! How were you able to get past the scans with missing data? Also, are the types of scans you mentioned sometimes harder to examine than others?

    ReplyDelete
  8. It's amazing to see you work around the difficulties and be able to analyze these patients. This is great training for the future. Keep it up!

    ReplyDelete
  9. Hi Richard! It is nice that you were able to work around all of the problems in finding the lesions, but how did you do so? Because if data is corrupted or missing, how do you still find the lesions? I can't wait for your next post!
    -Jileena

    ReplyDelete
  10. Hi Richard! It's great that you are able to familiarize yourself with these problems. I hope you make more progress and find more solutions to more problems

    ReplyDelete
  11. Looks like you're making great progress

    ReplyDelete
  12. Lol, instead of machine-learning, it's human-learning.
    Aside from that, this process sounds fairly prone to error. How are you isolating this version of liver cancer from other types. You don't seem to be able to verify whether something is actually HCC. Maybe I'm wrong.

    ReplyDelete

Post a Comment

Popular posts from this blog

Gathering the Data

Getting Started

Image Analysis And What I Learned