Mathew Leonardi | Searcharxiv

arXiv2026

EndoExtract: Co-Designing Structured Text Extraction from Endometriosis Ultrasound Reports

Endometriosis ultrasound reports are often unstructured free-text documents that require manual abstraction for downstream tasks such as analytics, machine learning model training, and clinical auditing. We present \textbf{EndoExtract}, an on-premise LLM-powered system that extracts structured data from these reports and surfaces interpretive fields for human review. Through contextual inquiry with research assistants, we identified key workflow pain points: asymmetric trust between numerical and interpretive fields, repetitive manual highlighting, fatigue from sustained comparison, and terminology inconsistency across radiologists. These findings informed an interface that surfaces only interpretive fields for mandatory review, automatically highlights source evidence within PDFs, and separates batch extraction from human-paced verification. A formative workshop revealed that \textbf{EndoExtract} supports a shift from field-by-field data entry to supervisory validation, though participants noted risks of over-skimming and challenges in managing missing data.

cs.HC↗

arXiv2026

Who Fails Where? LLM and Human Error Patterns in Endometriosis Ultrasound Report Extraction

In this study, we evaluate a locally-deployed large-language model (LLM) to convert unstructured endometriosis transvaginal ultrasound (eTVUS) scan reports into structured data for imaging informatics workflows. Across 49 eTVUS reports, we compared three LLMs (7B/8B and a 20B-parameter model) against expert human extraction. The 20B model achieved a mean accuracy of 86.02%, substantially outperforming smaller models and confirming the importance of scale in handling complex clinical text. Crucially, we identified a highly complementary error profile: the LLM excelled at syntactic consistency (e.g., date/numeric formatting) where humans faltered, while human experts provided superior semantic and contextual interpretation. We also found that the LLM's semantic errors were fundamental limitations that could not be mitigated by simple prompt engineering. These findings strongly support a human-in-the-loop (HITL) workflow in which the on-premise LLM serves as a collaborative tool, not a full replacement. It automates routine structuring and flags potential human errors, enabling imaging specialists to focus on high-level semantic validation. We discuss implications for structured reporting and interactive AI systems in clinical practice.

cs.HC↗