||Tram Nghi Pham
||University of California, Berkeley
||Spatiotemporal Big Data Analytics for Osteoarthritis Detection
The Osteoarthritis Initiative (OAI) is a multicenter, longitudinal, prospective, observational study of knee osteoarthritis launched by the National Institutes of Health in 2002. The OAI has accumulated a massive amount of clinical and imaging data and biological specimens from thousands of volunteers with risk factors for early knee osteoarthritis for a total of 8 years of follow up. However, these data have not been fully exploited to make optimal decision for the improvement of the prevention and intervention strategies of knee osteoarthritis.
This study focuses on spatiotemporal data analytics for osteoarthritis detection. The goal of this project is to identify clinical biomarkers for early detection of osteoarthrosis and improve the prevention and intervention strategies of knee osteoarthritis used in current clinical practice. Existing methods on spatiotemporal data analytics, dimensionality reduction, feature extraction and selection, and anomaly detection will be reviewed and studied.
- Week 1:
- This week, I spent most of my time reading research papers regarding my project.
We were able to narrow down some general methods which help me become familiar with the subject.
The main method, which is proposed by my mentor, will be studied carefully. I ended this week by finishing up
the presentation for next Monday and doing some image pre-processing implementations using Matlab.
- Week 2:
- I started working on method 1 - image processing to detect the boundary of cartilage.
I haven't finished the code. The intensity of each pixel oscillates so it is hard to determine
the threshold. I also read some papers regarding tensor decomposition, principal component analysis and its variants, etc.
to get myself familiar with method 2.
- Week 3:
- This week, I was able to detect the inner and outer boundary of cartilage after spending a descent amount of time
debugging the code that wasn't wrong. However, there are serveral images that need special treatment since it is really hard to determine
which part of the image is the cartilage. I read some articles about cartilage, OA knee in order to understand MRI images better.
Next week, I will spend time to improve my implementations.
- Week 4:
- I finished detecting the break point of cartilage for severe and minor OA knee using linear regression and least square
approximation. I tried to make my code efficient so that It works for all cases. We also decided to model the cartilage thickness in
3D using Matlab and calcualte its volume. I haven't done any 3D model simulations in Matlab, so it will be really cool to learn about
- Week 5:
- This week, I was trying to figure out how the volume of the cartilage can be calculated. I first explored 3-D Volumetric Data with
the Volume Viewer App which helped me to visualize my cartilage in 3D. I stacked 2D slices of gray image to create a 3D model then use it
as an input to Volume Viewer App. I rewrote the program by combining all individual functions so that it can run automatically. There was also a field trip to IBM this week.
- Week 6:
- Calculating the volume turned out to be more complicated than I thought. I run the program on several images and stacked the segmented images
to create a 3D model. We seperated the models of cartilage of OA and normal knees. I also recored the pixel coordinates of the cartilage of all images and viewed them in a sepreated 3D graph.
The volume then can be calcuated by interpolating between each 2D slices in 3D model or using the pixel coordinates of cartilages. I also worked on the final presentation which will be next Thrusday.
- Week 7:
- This week, I was able to segment the cartilage from the knee and sucessfully calculate the area of the cartilage.
After segmenting the cartilage, it became much more easier to model the cartilage in 3D and help to calculate its volume.
The thickness was sucessfully computed in the femur tibia regions. We also had a final 12-min presentation this week.
Next week, I will try to finish implementing the algorithms for 3D modeling and make the entire code more effective (since 3D modeling and segmentation will be added).
- Week 8:
- I finished cleaning up the code this week and continue working on calculating the volume of segmented cartilage.
I was able to compute the volume in the case where each layer of cartilage has the same area, but interpolation and other method will be utilized for other cases.
I ran some final experiments to test the code. The data is not complete yet so I tested my code on the sample data available on Osteoarthritis Initative.
I ended this week by starting my final report.
- Week 9:
- This week, I finished my final report and evaluation to the program. We are still waiting for data to be completed so I can run more experiements.
I will run more experiments to check the accuracy and consistence of the results. Once the data and experimental results are obtained, I will submit the paper and
we are hoping for a publishcation. We are all leaving this week. I will miss the program and everyone at DIMACS.