A major hindrance to quickly building models and reacting to new infectious diseases is the restricted availability of (quality) data. Developing machine learning models to assist clinical decision making in the beginning of a pandemic can be critical as it can shorten time-to-diagnosis and support specialized medical staff in an emergency setting 1. Having fast and reliable ways to screen infected patients is a learning from the COVID-19 pandemic. Utilizing this implicit data quality-quantity trade-off can be a promising approach for many scarce data domains beyond medical imaging. The best model of this study, called PneuKnowNet, shows an improvement of + 2.3% points in overall accuracy compared to a baseline model and also leads to more meaningful decision regions. We evaluate the proposed process on our study data for multiple types of models, all of which show improved results. This form of guidance by a human expert improves model convergence and performance. This process enables deep learning models to utilize annotated regions in the images during the training process. To successfully train a model on this small amount of complex image data, we propose a special knowledge distillation process, which we call Human Knowledge Distillation. The data was reviewed, differentiated into 4 causes for pneumonia, and annotated by an expert radiologist.

In this work we train a deep learning model on university hospital chest X-ray data, containing 1082 images. The successful training of deep learning models usually requires large amounts of quality data, which are often not available in many medical imaging tasks. In medical imaging, deep learning models can be a critical tool to shorten time-to-diagnosis and support specialized medical staff in clinical decision making.

Annotation journal