Extract Report Text

The canonical pipeline turning free-text pathology reports into structured data.

Purpose

Convert narrative report content coming out of the LIS into plain text and then structured fields, as the low-level layer beneath the LLM-based report-quality tooling.

Data used

Report corpora from the LIS (Patoloji Bilgi Yönetim Sistemi (LIS)) and outside-lab reports arriving through consultations (Konsultasyon). Historical corpora live in patolojirapor and patolojiraporlari. All research pulls pass through de-identification before touching a repo; research repos consume cleaned tables only.

Methods

extract-report-text handles PDF / HL7 / RTF to plain text. pathology-reports-text-analysis runs NLP over the extracted corpus. gleason_extraction is a worked example of extracting a specific structured field from narrative text. report-management-system sits alongside. Helpers: markitdown, repo2file.

Current state / open questions

Active. Onboarding first step: run the extractor on one de-identified PDF from the anonymized corpus and confirm the JSON output round-trips through the skills collection. The chapter itself is a stub — repository links only, no documented accuracy or evaluation. [unverified]

Open question: there is no recorded extraction-accuracy evaluation. For a pipeline feeding downstream cohort definitions this matters — extraction error propagates silently into every study that consumes the cleaned tables.

Derived from: text-analysis/extracttext.qmd, onboarding.qmd, lis/lis.qmd.