The purpose of this study is to analyze the efficacy of transfer learning\ntechniques and transformer-based models as applied to medical natural language\nprocessing (NLP) tasks, specifically radiological text classification. We used\n1,977 labeled head CT reports, from a corpus of 96,303 total reports, to\nevaluate the efficacy of pretraining using general domain corpora and a\ncombined general and medical domain corpus with a bidirectional representations\nfrom transformers (BERT) model for the purpose of radiological text\nclassification. Model performance was benchmarked to a logistic regression\nusing bag-of-words vectorization and a long short-term memory (LSTM)\nmulti-label multi-class classification model, and compared to the published\nliterature in medical text classification. The BERT models using either set of\npretrained checkpoints outperformed the logistic regression model, achieving\nsample-weighted average F1-scores of 0.87 and 0.87 for the general domain model\nand the combined general and biomedical-domain model. General text transfer\nlearning may be a viable technique to generate state-of-the-art results within\nmedical NLP tasks on radiological corpora, outperforming other deep models such\nas LSTMs. The efficacy of pretraining and transformer-based models could serve\nto facilitate the creation of groundbreaking NLP models in the uniquely\nchallenging data environment of medical text.\n