Natural Language Processing of Radiology Reports: Predicting Metastatic Progression from Text Data
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
The project’s goal is to extract tumour measurement data from oncological radiology reports with equal, or improved accuracy of a human radiologist. The purpose is to streamline and improve the efficiency of cancer diagnosis, accomplishing this through methods in artificial intelligence. In this experiment, a collection of 85,218 colorectal, and lung radiology reports were used. After loading reports into a data frame, and a BioBERT (bidirectional encoder representations from transformers pre-trained on biomedical corpora) model into a virtual environment, a question is set to be answered by the model where the context of the question is each radiology report’s findings section of the specified organ. These inputs are tokenized and embedded into numerical values to map sentences to vectors of real numbers. Vectors are fed into the model, and an answer is output in natural language for human readability. Answers are stored in the data frame in their corresponding row from which they were derived. The model successfully answered questions about measurements of tumours written in free text in reports where tumours were present, and successfully ignored or did not report in cases where tumours were not present, or measurements were unchanged. In cases with multiple tumours, the model reported exclusively first listed measurements. In an updated version of this model, context will be run through sentence-wise to ensure equal attention to the context entirely. This project is evidence that using one question and selecting the findings portion of a radiology report for one organ as context in a question-answering model built using BioBERT is effective, and efficient in collecting measurements from radiology reports. This algorithm can be applied to other areas of medicine, or other fields entirely with a few model alterations. This project is a step forward in improving cancer diagnosis efficiency and improving medical AI.
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 | 0.005 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.000 | 0.001 |
| 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