In the last years the amount of medical data and research procedures is rapidly increasing. With it there are even more possibilities for machine learning research, which has been present in the medical fields for a long while. Traditional approaches to data analysis have proven to be useful. The same goes for deep neural networks and more specifically for convolutional neural networks, which have seen heavily increased use in the latest years. They are mostly used on image data, although lately we have seen examples of them being used on conventional attribute data. We try to build a model based on CNN which would receive regular attribute data as input. We use a data set, which has many attributes (research results), many classes (different diseases) and many undefined values (different patients require different procedures, which results in many undefined attributes). It is on this real life data that we test different structures of deep and shallow CNN's. We conclude that results obtained in this manner are comparable to the best results on this data so far.
|
