Deep learning for medical report texts
Data storing within the medical sector is often stored as free texts entries, this is especially true for report texts, which are written after an examination. To be able to gather data from the report texts they need to be analyzed and classified to show what findings the examinations had. This thesis compares three state of the art deep learning approaches to classify the short medical report texts. This is done for two types of examinations, so the concept of transfer learning plays a role in the evaluation. An optimal model should learn concepts that are applicable for more than one type of examinations, since we can expect the report texts to be similar. The two dataset from the examinations are also of different sizes, and both have an uneven distribution among the target classes. One of the models is sequential and two of them convolutional, where one of the convolutional is a horizontal and one is vertical. The horizontal convolutional proves to be the best model across the different metrics, not the least in the sense of transfer learning as it improves its results when learning from both types of examinations. This hold true for all the models in regards of the lower represented classes as the transfer learning approach increases the accuracy of the models. All of the models behave similar to how a human reader might interpret the texts in terms of reliability and confidence.