MIMIC-III Label Extraction and My Journey in Omdena
I would like to give a quick brief on the challenge I was a part of. Our transformative challenge is to predict the occurrence of cardiac arrest using AI. Sudden Cardiac Arrest (SCA) is a medical emergency in which the heart suddenly stops beating, killing the patient within minutes. Survival rates for SCA are < 25% within hospitals. SCA can be prevented if the underlying cause is identified and treated. When SCA occurs, immediate electric shock to the heart improves chances of survival, but delays of just two minutes between SCA onset and shock lower survival rates and raise rates of brain damage among survivors. No technology on the market alerts clinicians of a patient’s risk of SCA. Current cardiac monitoring machines only issue alerts after a patient experiences an arrest, forcing healthcare providers to race against time.
The goal of this project is to expand the cardiac arrest prediction algorithm to pulseless electrical activity and asystole, providing an all-cause cardiac arrest prediction algorithm for more than 90% of patients. One of the parts of doing this involves label extraction and what has been done in it is explained further. The data was majorly sourced from the MIMIC III Clinical database.
In the label extraction task, we segregate data based on various parameters to determine if the person had a cardiac arrest or not. We had labels from two distinct sets, namely Sepsis labeled data and Transformative labeled data for the same records. These datasets have a cardiac arrest time or mention if there is a cardiac arrest without charting the time or can be null. We categorize the combinations based on null values, cardiac arrest, and NOT CHARTED data into 8 categories, namely:
- Sepsis is null, Transformative has a Cardiac Arrest time.
- Sepsis is null, Transformative has Cardiac Arrest but they have not charted time.
- Sepsis is null, Transformative is null.
- Sepsis is null, Transformative has Cardiac Arrest time but does not match the description extracted from the data.
- Sepsis has cardiac arrest time, Transformative is null.
- Sepsis has cardiac arrest time, Transformative has a Cardiac Arrest time.
- Sepsis has cardiac arrest time, Transformative has Cardiac Arrest but they have not charted time.
- Sepsis has cardiac arrest time, Transformative has Cardiac Arrest time but does not match the description extracted from the data.
This labeling task involves verifying the records under any of the eight combinations mentioned above to determine the correctness based on the text extracted from the records. We determine if the person had a cardiac arrest (Agree) or not had an arrest (disagree) or sometimes it’s neutral (cardiac arrest is likely but uncertain).
This labeling helps us to correctly classify the data and provide new insights to improve the model’s performance.
Apart from this, My journey in Omdena in this transformative project has been fascinating. I worked on various tasks like exploratory analysis, label investigation & data gathering stage to work with MIMIC data to find the potential features from the paper and online resources. I have learned a lot of new information after completing the challenge and it was a great learning experience. Special thanks to Sanjana Tule for her guidance and Sijuade Oguntayo for his support.
Also, Omdena projects usually have a systematic approach to track the deadlines and monitor the work done by us, encouraging us to do coding sprints, support our ideas and make it work by providing full support. There is no hierarchy and anyone can pick up anything to work upon. The constant support by the project manager and follow-ups is what makes these projects end with fruitful work. You have to put your efforts to be a part of the project. The most fascinating thing is they monitor each collaborator’s weekly progress and provide feedback and push us to work in a healthy, motivating environment with collaborators all over the world, which has a huge positive impact left on us. This promotes our leadership skills, communication skills, and positive competition which can be applied in the real world. Last but not least, you get a community that you can cherish and contact for help at any time and this is real. This has been my wonderful experience so far with Omdena.
