Derivation and validation of text search algorithms for renal and adrenal lesion identification in radiology text reports
Bibliographic record
Abstract
Introduction: Most cohort studies are limited by sampling and accrual bias. The capability to detect specific lesions identified in radiological text reports could eliminate these biases and benefit patient care, clinical research, and trial recruitment. This study derived and internally validated text search algorithms to identify four common urological lesions (solid renal masses, complex renal cysts, adrenal masses, and simple renal cysts) using radiology text reports. Methods: A simple random sample of 10 000 abdominal ultrasound (US) and computed tomography (CT) reports was drawn from our hospital’s data warehouse. Reports were manually reviewed to determine the true status of the four lesions. Using commonly available software, we created logistic regression models having as predictors the status of a priori selected text terms in the report. We used bootstrap sampling with 95th percentile thresholds to select variables for the final models, which were modified into point systems. A second independent, random sample of 2855 reports, stratified by the number of points for each abnormality, was reviewed in a blinded fashion to measure the accuracy of each lesion’s point system. Results: The prevalence of solid renal mass, complex renal cyst, adrenal mass and simple renal cyst, was 2.0%, 1.7%, 3.2%, and 20.0%, respectively. Each model contained between one and five text terms with c-statistics ranging between 0.66 and 0.90. In the independent validation, the scoring systems accurately predicted the probability that a text report cited the four lesions. Conclusions: Textual radiology reports can be analyzed using common statistical software to accurately determine the probability that important abnormalities of the kidneys or adrenal glands exist. These methods can be used for case identification or epidemiological studies.
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 | 0.086 | 0.346 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.016 | 0.006 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.005 | 0.003 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.002 | 0.001 |
| 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".