Natural Language Processing of Large-Scale Structured Radiology Reports to Identify Oncologic Patients With or Without Splenomegaly Over a 10-Year Period
Bibliographic record
Abstract
PURPOSE To assess the accuracy of a natural language processing (NLP) model in extracting splenomegaly described in patients with cancer in structured computed tomography radiology reports. METHODS In this retrospective study between July 2009 and April 2019, 3,87,359 consecutive structured radiology reports for computed tomography scans of the chest, abdomen, and pelvis from 91,665 patients spanning 30 types of cancer were included. A randomized sample of 2,022 reports from patients with colorectal cancer, hepatobiliary cancer (HB), leukemia, Hodgkin lymphoma (HL), and non-HL patients was manually annotated as positive or negative for splenomegaly. NLP model training/testing was performed on 1,617/405 reports, and a new validation set of 400 reports from all cancer subtypes was used to test NLP model accuracy, precision, and recall. Overall survival was compared between the patient groups (with and without splenomegaly) using Kaplan-Meier curves. RESULTS The final cohort included 3,87,359 reports from 91,665 patients (mean age 60.8 years; 51.2% women). In the testing set, the model achieved accuracy of 92.1%, precision of 92.2%, and recall of 92.1% for splenomegaly. In the validation set, accuracy, precision, and recall were 93.8%, 92.9%, and 86.7%, respectively. In the entire cohort, splenomegaly was most frequent in patients with leukemia (32.5%), HB (17.4%), non-HL (9.1%), colorectal cancer (8.5%), and HL (5.6%). A splenomegaly label was associated with an increased risk of mortality in the entire cohort (hazard ratio 2.10; 95% CI, 1.98 to 2.22; P < .001). CONCLUSION Automated splenomegaly labeling by NLP of radiology report demonstrates good accuracy, precision, and recall. Splenomegaly is most frequently reported in patients with leukemia, followed by patients with HB.
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 | 0.005 | 0.024 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| 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 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".