Original title: Bridging the data latency gap: automated extraction of genomic biomarkers from unstructured clinical documents to support real-world oncology data
Qianyun Luo, Rui Zhang, Nikitha Vobugari, Jane Y. C. Hui +1 · September 2026
Doctors need cancer test results quickly to choose the best treatments and run studies. However, important gene test scores often get trapped inside scanned paper documents. Right now, workers must read these files and type the numbers into databases by hand, which creates long delays. To solve this, researchers tested three free computer vision tools that read text from images. They used these tools to pull cancer recurrence scores out of scanned reports and compared the automated reads against human reviews. The tools work by scanning the document image, identifying the printed letters and numbers, and lifting the specific score directly from the page. A combined setup that paired two tools worked best. It matched human accuracy almost perfectly while slashing the time spent reading documents. Automating this work speeds up access to cancer data, helping modernize registries without relying on slow manual entry.
Imagine you receive a handwritten list of ingredients from a friend. You have to read it and type it into your phone. But the handwriting is messy, and sometimes you skip a word or misread a number. Now imagine you have two friends who can read the list, but each has different strengths: one is good at reading cursive, the other at numbers. By asking both friends to read it and then combining their answers, you get the full correct list almost every time. In this paper, the 'list' is a scanned cancer test report, the 'friends' are different OCR programs, and combining their outputs—a hybrid approach—pulls out the test scores with near-perfect accuracy.
Cancer registries rely on manual abstraction of genomic data from scanned documents, causing delays in research and patient care. Automating this step could speed up the availability of key biomarkers like Oncotype DX scores, improving real-world evidence generation and enabling faster treatment decisions. This study shows that a hybrid OCR approach can achieve human-level accuracy while cutting processing time.
The study was conducted at a single Midwestern U.S. health system with a specific marker (Oncotype DX). Generalizability to other institutions, document formats, or biomarkers is not established. The hybrid approach may require tuning for different report layouts. Unknown PR status was the only factor predicting registry discordance, but other unmeasured variables may exist.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Bridging the data latency gap: automated extraction of genomic biomarkers from unstructured clinical documents to support real-world oncology data — Qianyun Luo, Rui Zhang, Nikitha Vobugari et al. — Cancer Causes & Control — doi:10.1007/s10552-026-02252-y
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.