Week 9: Finalizing Patient Detection

Apr 24, 2021

Hi all! Welcome back to my blog for week 9. In today’s blog, we will cover my progress this week as well as discussing the end of my work on patient detection with future steps. As you may have already noticed, this blog will be much shorter than my usual blogs since there is less technical material to cover from this week in particular. For a recap of last week, we discussed my progress towards post-processing techniques and finalizing some technical steps. To summarize my work this week, I have started collecting my results to create graphs and write a paper! Let’s get into this below:

Writing a Paper:

Since I am finished with the majority of the technical work in my project, it makes sense to wrap up my work and present it in a paper format. This is preferable since the paper would provide a deeper dive than my presentation and I could also virtually distribute my paper at the senior project presentations. My current goal is to have a 3-5 paper that captures the essence of my project. Here is my current planned outline:

  • Introduction (0.5 page)
    • Go over the main ideas of Video EEG 
    • Discuss problem
    • Briefly cover object detection
  • Related Work (1 page)
    • Reference existing paper
    • Cover more technical details on Object Detection and Person Tracking
    • Discuss IoU and bounding boxes
  • Methods (3-4 pages)
    • Preprocessing (1-2 pages)
    • Object Detection (2-3 pages)
    • Post Processing (1 page)
  • Results (1 page)
    • Cover results (Average Precision, etc.)
    • Show graphs for fine-tuned model vs no fine-tuned model
    • Wrap up with the main contribution

As you can see, this outline is pretty vague and only mentions small details about each section. This week, I began making progress on this paper and finished the introduction and abstract sections. The abstract is similar to my senior project abstract but also includes the numerical results. The overall purpose of this paper is to show the application of object detection towards EEG patient detection and tracking, while also having reproducible results which can set the foundation for future work in this new domain. I will also demonstrate how fine-tuning the Detectron2 model on a set of 30 videos enables better performance than existing pre-trained models. I aim to have multiple graphs to accompany these discussions of my algorithm. 


In conclusion, we covered details on the finalizing of patient detection as well as my progress in terms of writing a paper. My goal is to share a draft of this paper in next week’s blog so stay tuned for that. Thanks for sticking through this blog!


Leave a Reply

Your email address will not be published.