Höfundar:
Chadi Barakat, Sebastian Fritsch, Morris Riedel
Many machine learning-based models have been developed with the aim of providing diagnosis or decision support in medical applications; these models are especially helpful in the time-critical situations often occurring in Intensive Care Units (ICUs). One specific condition that affects may ICU patients with a high mortality rate is Acute Respiratory Distress Syndrome (ARDS), which is heterogeneous, difficult to diagnose, and difficult to treat. This research takes advantage of High-Performance Computing resources and efficient data processing and machine learning algorithms to convert a pre-exisitng simulator into a diagnostic model for predicting onset of ARDS. The developed model performance closely mimics the performance of the original simulator it is based on, with a minimum R2 score of .92 per predicted parameter. Through the development and application of models like the one presented in this research, diagnosis can be accelerated leading to improved outcomes for patients.
