Using a Longformer Large Language Model for Segmenting Unstructured Cancer Pathology Reports
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
PURPOSE: Many Natural Language Processing (NLP) methods achieve greater performance when the input text is preprocessed to remove extraneous or unnecessary text. A technique known as text segmentation can facilitate this step by isolating key sections from a document. Give that transformer models-such as Bidirectional Encoder Representations from Transformers (BERT)-have demonstrated state-of-the-art performance on many NLP tasks, it is desirable to leverage such models for segmentation. However, transformer models are typically limited to only 512 input tokens and are not well suited for lengthy documents such as cancer pathology reports. The Longformer is a modified transformer model designed to intake longer documents while retaining the positive characteristics of standard transformers. This study presents a Longformer model fine-tuned for cancer pathology report segmentation. METHODS: We fine-tuned a Longformer Question-Answer (QA) model on 504 manually annotated pathology reports to isolate sections such as diagnosis, addenda, and clinical history. We compared baseline methods including regular expressions (regex) and BERT QA. However, those methods may fail to correctly identify section boundaries. Model performance was evaluated using sequence recall, precision, and F1 score. RESULTS: Final test results were obtained on a hold-out test set of 304 cancer pathology reports. We report sequence F1 scores for the following sections: diagnosis (0.77), addenda (0.48), clinical history (0.89), and overall (0.68). CONCLUSION: We present a fine-tuned Longformer model to isolate key sections from cancer pathology reports for downstream analyses. Our model performs segmentation with greater accuracy.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it