Although the data volume of medical imaging is not large and privacy must be considered, because it is a low-context related problem, it is an area where machine learning can play a big role.However, Shangzhizhi initially encountered the challenge of "insufficient data". Although it has accumulated about 2000~3000 ECG data for nearly seven years, it still belongs to "small batch data", which is not enough to achieve machine.
Learning model calculation, so the initial data processing accuracy is only 50%. Fortunately, Amazon SageMaker provides data resources and data scientists assistance, and then through data model conversion, the data owned by Shangzhiji is expanded whatsapp list by 10 times. Even if there are many ECG patterns, the accuracy can exceed 80%. Not only that, the expansion from the originally familiar medical images to the interpretation of abnormal ECGs seems to belong to the medical field, but the service design is very different.