Clinically Accurate Report Generation

There has been an impressive amount of progress with generating descriptive reports from radiographs, employing state-of-the-art convolutional neural networks, and complex Natural Language Processing. However, there is a gap between current models and ingrained nuances of the radiology domain that the models fail to leverage. This project explores this gap and finds possible solutions by enhancing the model architecture, extracting radiomic features from the radiographs, and generating an end-to-end architecture that employs radiomic feature extraction, a CNN encoder, and an RNN sentence and word decoder to generate clinically accurate reports from radiographs.