Medical Analysis of Uveitis Patients
During the spring semester of 2021, I was part of a group semester project analyzing medical patient data. The goal of the project was to explore whether ocular disease, specifically uveitis, can be predicted by a variety of attributes such as blood and inflammation values. Causes of this disease are wide ranging and have not yet been fully explored. The results of our analysis should therefore help to identify potential patterns in the existing data and thereby contribute to a better understanding of the disease.
A total of four students had signed up for this project. However, we decided to form two teams of two, since we considered the overall control and the individual learning impact to be higher with this group size. I was very excited about the challenge because it was the first larger opportunity for me to implement machine learning algorithms at a real-world setting.
At the beginning, we had to clean the data and bring it into a suitable format. Since we ourselves did not have any medical know-how, a close exchange with the experts was necessary for this step. Because we only had measurements from nearly 1100 patients, it was also a challenge to make significant predictions, especially for classifications with many different labels. In the beginning, we worked more closely together, forming a common understanding of the data. Later, we divided the algorithms to be applied equally among each other and focused on different parts of the project, allowing for a more independent way of working.
Overall, the cooperation and distribution of workload in the group worked very smooth. I think one reason for the successful cooperation is that all team members already knew each other and had worked together before. As a result, there was a good collaboration in the team right from the start and we were able to distribute the tasks in a way that the individual group members benefit the most for their learning progress. We also had a lot of support from our medical experts’ team. Without the appropriate medical know-how, it would have been much more difficult for us to interpret our results and extract the full information value. The workload was well distributed.
In conclusion, I can say that the project was very helpful as an addition and extension to my theoretical knowledge about basic machine learning. It was very interesting to see how each machine learning algorithm performed on our dataset compared to each other. In addition, it was very rewarding for me to be able to work with real external partners who, through their specific expertise and demands, significantly enhanced the outcome of our project.