Overview
In the diagnostic pursuit of rare genetic and metabolic disorders, subjective clinical notes must invariably be corroborated by objective laboratory data. A physician’s observation of muscle weakness is crucial, but a serum creatine kinase level of 10,000 U/L is an undeniable, mathematically precise biological truth. However, the mechanism by which this critical laboratory data is transferred between disconnected healthcare systems remains archaic. When a patient arrives at a specialist clinic, their historical lab results are rarely transmitted as structured, interoperable data packets (like FHIR or HL7 v2 messages). Instead, they almost always arrive as flattened, multi-page PDF documents or degraded paper faxes. For standard Natural Language Processing (NLP) engines, extracting information from these flattened laboratory PDFs is notoriously difficult. Standard text-extraction algorithms read left-to-right, top-to-bottom, completely destroying the visual, columnar layout of a lab report table. If a PDF contains an analyte name in column one, the observed value in column two, and the reference range in column three, a rudimentary text parser will scramble these figures into an incomprehensible paragraph, completely severing the link between the biological marker and its corresponding quantitative result. Sanjeevni specifically engineered a proprietary Table Parsing Engine to overcome this precise infrastructural failure. When a laboratory PDF is uploaded into the system, our advanced computer vision algorithms do not merely read the text; they mathematically map the geometric structure of the document. The AI identifies the vertical and horizontal boundaries of the tables, correctly delineating the headers (Analyte, Result, Units, Reference Range). This allows Sanjeevni to accurately reconstruct the tabular data perfectly within our secure database, ensuring that a critical potassium value is never accidentally attributed to a sodium reference range.
The Challenge of Ingesting Tabular Data
Laboratory values are the objective truth. However, their transmission as static PDFs creates a massive data entry bottleneck. Sanjeevni solves this with advanced computer vision.
Automated Abnormality Flagging and HPO Mapping
Lumina automates the identification of laboratory abnormalities, mapping them directly to standardized phenotypic terms to accelerate the diagnostic process.
Preventing Transcription Errors and Data Silos
Manual data entry of laboratory values is a persistent source of medical error. Sanjeevni eliminates this risk by digitizing the process directly from source records.