Extract data from pathology reports

A pathology report is the specimen-level record a pathologist signs after examining tissue removed from a patient, and it is the document that names the disease. A surgical pathology report on a breast core needle biopsy carries an accession number such as S26-04412, the ordering physician Dr. Priya Nair, the signing pathologist Dr. Alan Whitfield, and three layers of examination: the gross description of the specimen as it arrived, the microscopic findings read under the slide, and the final diagnosis. Tumor registrars, oncology coordinators, and cancer center data teams read these reports to register cases, stage disease, and populate the electronic medical record (EMR), each report tying back to a patient by medical record number and date of birth. The difficulty sits in the ancillary studies and the coded diagnosis. A breast carcinoma report rarely stops at the narrative: it carries an immunohistochemistry (IHC) panel reporting estrogen receptor, progesterone receptor, HER2, and Ki-67 results, sometimes special histochemical stains, and molecular studies, each with its own antibody or marker and a positive or negative outcome. The diagnosis line pairs prose, invasive ductal carcinoma of Nottingham grade 2 under the World Health Organization classification of tumors, with a coded morphology under ICD-O, and a report can be amended or issued as an addendum dated 2026-06-30 once the stains resolve. College of American Pathologists synoptic templates push some of this into checklists, yet most reports still mix free text with a study grid. Talonic reads the pathology report into a structured record that keeps the specimen, the diagnosis, and the ancillary studies apart. A final report dated 2026-06-26 on a right breast core biopsy collected 2026-06-24 for patient Margaret Halloran at Mercy Regional Laboratory returns its gross description, its microscopic findings, and a diagnosis of invasive ductal carcinoma coded 8500/3, with the ER, PR, HER2, and Ki-67 results in a studies table, so a tumor registry or an EHR works from fields rather than a scanned report. The report is protected health information (PHI), and Talonic returns the pathologist's stated findings and diagnosis exactly as written, without interpreting the slides or reaching a diagnosis of its own.

What gets extracted from pathology reports

How extraction works for pathology reports

Pathology reports come off laboratory information systems in surgical pathology, cytology, and hematopathology layouts, so the gross, microscopic, and diagnosis sections and the placement of the study grid shift from one lab to the next. Talonic classifies the report and maps it to the clinical schema in the Field Registry, which separates the specimen and its collection detail from the microscopic findings and from the coded diagnosis. Special stains, immunohistochemistry markers, and molecular studies each return as a row in a studies table with their result, and the diagnosis carries its ICD-O morphology code where the report states one. Every value returns with a confidence score and pixel-region provenance under DIN SPEC 91491 conformity, and because the report is protected health information, the fields are released only to the account that submitted it. Talonic returns the findings and the diagnosis as written and does not read the slides or make a clinical determination.

Sample extraction

A final surgical pathology report on a breast core needle biopsy

{
  "document_number": "S26-04412",
  "document_date": "2026-06-26",
  "status": "final",
  "patient.name": "Margaret Halloran",
  "patient.medical_record_number": "00814423",
  "patient.date_of_birth": "1968-11-02",
  "patient.gender": "female",
  "ordering_provider.name": "Dr. Priya Nair",
  "pathologist.name": "Dr. Alan Whitfield",
  "specimen.type": "Core needle biopsy",
  "specimen.collection_date": "2026-06-24",
  "specimen.anatomic_location": "Right breast, upper outer quadrant",
  "specimen.laterality": "right",
  "clinical_history": "52-year-old with a suspicious mass on screening mammography, BI-RADS 4",
  "gross_description": "Three cores of tan-white fibrofatty tissue, aggregate 2.1 cm, entirely submitted in cassette A1",
  "microscopic_findings": "Infiltrating nests and cords of atypical epithelial cells within a desmoplastic stroma; no in-situ component identified in these cores",
  "diagnosis": "Invasive ductal carcinoma, Nottingham grade 2",
  "immunohistochemistry": "ER positive (95%), PR positive (80%), HER2 negative (score 1+), Ki-67 15%",
  "special_stains": "None performed",
  "molecular_studies": "Deferred to excision specimen",
  "conclusion": "Findings consistent with invasive ductal carcinoma; recommend correlation with imaging and surgical planning",
  "report_category": "Surgical pathology",
  "performer": "Mercy Regional Laboratory",
  "specimens": [
    {
      "specimen_id": "A1",
      "type": "Core needle biopsy",
      "collection_date": "2026-06-24",
      "anatomic_location": "Right breast, upper outer quadrant",
      "laterality": "right"
    }
  ],
  "diagnoses": [
    {
      "diagnosis_id": "D1",
      "diagnosis": "Invasive ductal carcinoma, Nottingham grade 2",
      "icd_code": "ICD-O 8500/3"
    }
  ],
  "special_studies": [
    {
      "study_id": "S1",
      "study_type": "immunohistochemistry",
      "study_name": "Estrogen receptor (ER)",
      "result": "Positive, 95%"
    },
    {
      "study_id": "S2",
      "study_type": "immunohistochemistry",
      "study_name": "Progesterone receptor (PR)",
      "result": "Positive, 80%"
    },
    {
      "study_id": "S3",
      "study_type": "immunohistochemistry",
      "study_name": "HER2",
      "result": "Negative, score 1+"
    },
    {
      "study_id": "S4",
      "study_type": "immunohistochemistry",
      "study_name": "Ki-67",
      "result": "15%"
    }
  ]
}

Frequently asked