HealthDataLabellingSystem | Data tagging system where various labels

HealthDataLabellingSystem is a Jupyter Notebook library. HealthDataLabellingSystem has no bugs, it has no vulnerabilities and it has low support. You can download it from GitHub.

As the description of the challenge states “More and more healthcare data is being generated by an exponentially increasing number of medical devices and applications.”. Healthcare system is usually on the bottom section of any government’s budget. Hospitals struggle from underfunding. Doctors and nurses are overwhelmed. The COVID-19 pandemic gives a lot of extra pressure on the healthcare system and on hospitals’ staff as well. There are a lot of standards in different fields of healthcare like the Dicom file format in radiology. Companies that operate near the healthcare industry, usually provide data via API services, but those APIs are not synchronized, they do not use the same labeling and tagging method. That’s why using that data is heavy and expensive. Doctors or third party companies have no deep programming knowledge, so if the hospitals have well trained IT staff, they can access data easily.We have personal experiences, since we tried smart bracelets already. When we went to a doctor to show the result, they didn’t know what to do with the exported data file. A properly labelled data stream with a useful UI provides accessibility for users who do not have IT skills. Companies could not solve the problem of uniformization since they are competitors and it is bad for the brand if their data labelling method is not chosen for the standard. From the other side, the company that provides the method for standardization has a great advantage over others. Nowadays everybody talks about big data and AI systems. At the same time, a lots of data in healthcare are still unlabeled. To train a better AI network, we should provide more and more data. Besides the quantity, the quality is also important. A rich labelled data helps to process the old data as well. For example a well trained NLP solution can process those documents better, that was written and stored before our labelling method. For example, one of the most famous datasets in healthcare related deep learning research is the NIH Chest X-ray Dataset made by the National Institutes of Health. It contains 14 different diseases and 1 label for negative cases. The creators use* Natural Language Processing to text-mine disease classifications from the associated radiological reports. The labels are expected to be >90% accurate. This seems a quite accurate dataset, but it can be better. Besides this, the biggest limitation is that the original associated radiological reports are not publicly available. That’s why we cannot make a cross-validation process or get a second opinion about data. This is a tricky question, since most of the people say data labelling has no real and present effect on an average human’s daily life. This is incorrect. As an average human, we are visiting doctors and hospitals as others. As we mentioned earlier, we had experience when a doctor could not read the exported data of a smart bracelet. We have a lot of personal health data from smart devices to the official health documents that cannot be merged into a whole dataset, since there is no proper labelling system yet. We could manage each type of data independently from others with the existing softwares. It is kind of funny and sad at the same time, but we have 5 different smart bracelets or watches and we have to use 5 different software or APIs. We are deep learning developers, so we struggle daily with small and poor quality datasets. We had to code a lot to implement an API solution into our workflow. So the problem impact on us not just as normal people, but as a professional as well. We are committed to healthcare and we like hackathons. Since we are data scientists or deep learning developers, a lot of experience was gained during the years. In the last year we won a hackathon with a healthcare related idea: a file format, which is called DoF*, that helps to store and share datasets easily. Originally we are lawyers, so we can solve the legal issues such as meeting the regulation of GDPR, HIPAA or CCPA. DoF is now integrated into Seagate’s CORTX framework. We think, we can add new viewpoints to solve the problem of this challenge and we have the necessary skills to build something useful, feasible and sustainable.