Machine-based Quantitative Scoring

The results of our model are very encouraging. Its scores are unbiased and in most cases very close to the ground truth. However, the results for some examples are still considerably off, calling for more research. In the present proposal the overarching goal is to further improve our machine-based scoring system in order to reach a prediction accuracy and functionality that allows us to bring a corresponding web application onto the market.

In our current BRIDGE Discovery project, we aim to further improve data preprocessing, data augmentation, and optimize neural networks to increase predictive accuracy, in order to meet the criteria of medical regulations to implement a valid scoring system. Furthermore we plan to add a quantitative uncertainty measure to the neural network to increase the practical usability of our scoring system. Finally, we develop a web application, in which users can upload the ROCF images and receive the ROCF scores instantly. We have developed a prototype for web-based scoring application.