Transformer versus traditional natural language processing: how much data is enough for automated radiology report classification?
Bibliographic record
Abstract
OBJECTIVES: Current state-of-the-art natural language processing (NLP) techniques use transformer deep-learning architectures, which depend on large training datasets. We hypothesized that traditional NLP techniques may outperform transformers for smaller radiology report datasets. METHODS: We compared the performance of BioBERT, a deep-learning-based transformer model pre-trained on biomedical text, and three traditional machine-learning models (gradient boosted tree, random forest, and logistic regression) on seven classification tasks given free-text radiology reports. Tasks included detection of appendicitis, diverticulitis, bowel obstruction, and enteritis/colitis on abdomen/pelvis CT reports, ischemic infarct on brain CT/MRI reports, and medial and lateral meniscus tears on knee MRI reports (7,204 total annotated reports). The performance of NLP models on held-out test sets was compared after training using the full training set, and 2.5%, 10%, 25%, 50%, and 75% random subsets of the training data. RESULTS: In all tested classification tasks, BioBERT performed poorly at smaller training sample sizes compared to non-deep-learning NLP models. Specifically, BioBERT required training on approximately 1,000 reports to perform similarly or better than non-deep-learning models. At around 1,250 to 1,500 training samples, the testing performance for all models began to plateau, where additional training data yielded minimal performance gain. CONCLUSIONS: With larger sample sizes, transformer NLP models achieved superior performance in radiology report binary classification tasks. However, with smaller sizes (<1000) and more imbalanced training data, traditional NLP techniques performed better. ADVANCES IN KNOWLEDGE: Our benchmarks can help guide clinical NLP researchers in selecting machine-learning models according to their dataset characteristics.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.010 | 0.039 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.000 | 0.002 |
| Scholarly communication | 0.003 | 0.010 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".