Share via


Azure Cognitive Services Health Insights Cancer Profiling-Clientbibliothek für Python– Version 1.0.0b1

Health Insights ist ein Azure Applied AI Service, der mit dem Azure Cognitive Services-Framework erstellt wurde und mehrere Cognitive Services, Healthcare API-Dienste und andere Azure-Ressourcen nutzt.

Das Krebsprofiling-Modell empfängt klinische Datensätze von Onkologie-Patienten und gibt Krebs-Staging aus, wie z. B. TNM-Kategorien im klinischen Stadium und TNM-Kategorien im pathologienischen Stadium sowie Tumorstandort, Histologie.

Quellcode | Paket (PyPI) | API-Referenzdokumentation | Produktdokumentation | Proben

Erste Schritte

Voraussetzungen

  • Für die Verwendung dieses Pakets ist Python 3.7 oder höher erforderlich.
  • Sie benötigen ein Azure-Abonnement , um dieses Paket verwenden zu können.
  • Eine vorhandene Cognitive Services Health Insights-instance.

Installieren des Pakets

pip install azure-healthinsights-cancerprofiling

Diese Tabelle gibt Aufschluss über die Beziehung zwischen SDK-Versionen und unterstützten API-Versionen des Diensts:

SDK-Version Unterstützte API-Version des Diensts
1.0.0b1 2023-03-01-preview

Authentifizieren des Clients

Abrufen des Endpunkts

Sie finden den Endpunkt für Ihre Health Insights-Dienstressource über das Azure-Portal oder die Azure CLI.

# Get the endpoint for the Health Insights service resource
az cognitiveservices account show --name "resource-name" --resource-group "resource-group-name" --query "properties.endpoint"

Abrufen des API-Schlüssels

Sie können den API-Schlüssel aus der Health Insights-Dienstressource im Azure-Portal abrufen. Alternativ können Sie den folgenden Azure CLI-Codeausschnitt verwenden, um den API-Schlüssel Ihrer Ressource abzurufen.

az cognitiveservices account keys list --resource-group <your-resource-group-name> --name <your-resource-name>

Erstellen eines CancerProfilingClient mit API-Schlüsselanmeldeinformationen

Sobald Sie über den Wert für den API-Schlüssel verfügen, können Sie ihn als Zeichenfolge an eine instance von AzureKeyCredential übergeben. Verwenden Sie den Schlüssel als Anmeldeinformationsparameter, um den Client zu authentifizieren:

import os
from azure.core.credentials import AzureKeyCredential
from azure.healthinsights.cancerprofiling.aio import CancerProfilingClient

KEY = os.environ["HEALTHINSIGHTS_KEY"]
ENDPOINT = os.environ["HEALTHINSIGHTS_ENDPOINT"]

cancer_profiling_client = CancerProfilingClient(endpoint=ENDPOINT, credential=AzureKeyCredential(KEY))

Long-Running Vorgänge

Vorgänge mit langer Ausführungsdauer sind Vorgänge, die aus einer anfänglichen Anforderung bestehen, die an den Dienst gesendet wird, um einen Vorgang zu starten, gefolgt von der Abfrage des Diensts in Intervallen, um festzustellen, ob der Vorgang abgeschlossen oder fehlgeschlagen ist, und ob er erfolgreich war, um das Ergebnis zu erhalten.

Methoden, die die Analyse im Gesundheitswesen, die benutzerdefinierte Textanalyse oder mehrere Analysen unterstützen, werden als zeitintensive Vorgänge modelliert. Der Client macht eine begin_<method-name> Methode verfügbar, die ein Abfrageobjekt zurückgibt. Aufrufer sollten warten, bis der Vorgang abgeschlossen ist, indem sie für das Von der begin_<method-name> -Methode zurückgegebene Poller-Objekt aufrufenresult(). Beispielcodeausschnitte werden bereitgestellt, um die Verwendung von Vorgängen mit langer Ausführungszeit zu veranschaulichen unten.

Wichtige Begriffe

Mit dem Krebsprofilerstellungsmodell können Sie Aus unstrukturierten klinischen Dokumenten Krebsattribute wie Tumorstandort, Histologie, TNM-Kategorien im klinischen Stadium und TNM-Kategorien im pathologienischen Stadium ableiten.

Beispiele

Der folgende Abschnitt enthält mehrere Codeausschnitte, die einige der häufigsten Health Insights – Cancer Profiling-Dienstaufgaben behandeln, einschließlich:

Krebsprofilerstellung

Schließen Sie wichtige Krebsattribute wie Tumorstelle, Histologie, TNM-Kategorien im klinischen Stadium und TNM-Kategorien im pathologischen Stadium aus den unstrukturierten klinischen Dokumenten eines Patienten ab.

import asyncio
import os
import datetime
from azure.core.credentials import AzureKeyCredential
from azure.healthinsights.cancerprofiling.aio import CancerProfilingClient
from azure.healthinsights.cancerprofiling import models

KEY = os.environ["HEALTHINSIGHTS_KEY"]
ENDPOINT = os.environ["HEALTHINSIGHTS_ENDPOINT"]

# Create an Onco Phenotype client
# <client>
cancer_profiling_client = CancerProfilingClient(endpoint=ENDPOINT,
                                                credential=AzureKeyCredential(KEY))
# </client>

# Construct patient
# <PatientConstructor>
patient_info = models.PatientInfo(sex=models.PatientInfoSex.FEMALE, birth_date=datetime.date(1979, 10, 8))
patient1 = models.PatientRecord(id="patient_id", info=patient_info)
# </PatientConstructor>

# Add document list
# <DocumentList>
doc_content1 = """
            15.8.2021
            Jane Doe 091175-8967
            42 year old female, married with 3 children, works as a nurse
            Healthy, no medications taken on a regular basis.
            PMHx is significant for migraines with aura, uses Mirena for contraception.
            Smoking history of 10 pack years (has stopped and relapsed several times).
            She is in c/o 2 weeks of productive cough and shortness of breath.
            She has a fever of 37.8 and general weakness.
            Denies night sweats and rash. She denies symptoms of rhinosinusitis, asthma, and heartburn.
            On PE:
            GENERAL: mild pallor, no cyanosis. Regular breathing rate.
            LUNGS: decreased breath sounds on the base of the right lung. Vesicular breathing. 
                No crackles, rales, and wheezes. Resonant percussion.
            PLAN:
            Will be referred for a chest x-ray.
            ======================================
            CXR showed mild nonspecific opacities in right lung base.
            PLAN:
            Findings are suggestive of a working diagnosis of pneumonia. The patient is referred to a
            follow-up CXR in 2 weeks."""

patient_document1 = models.PatientDocument(type=models.DocumentType.NOTE,
                                           id="doc1",
                                           content=models.DocumentContent(
                                               source_type=models.DocumentContentSourceType.INLINE,
                                               value=doc_content1),
                                           clinical_type=models.ClinicalDocumentType.IMAGING,
                                           language="en",
                                           created_date_time=datetime.datetime(2021, 8, 15))

doc_content2 = """
            Oncology Clinic
            20.10.2021
            Jane Doe 091175-8967
            42-year-old healthy female who works as a nurse in the ER of this hospital.
            First menstruation at 11 years old. First delivery- 27 years old. She has 3 children.
            Didn't breastfeed.
            Contraception- Mirena.
            Smoking- 10 pack years.
            Mother- Belarusian. Father- Georgian. 
            About 3 months prior to admission, she stated she had SOB and was febrile.
            She did a CXR as an outpatient which showed a finding in the base of the right lung-
            possibly an infiltrate.
            She was treated with antibiotics with partial response.
            6 weeks later a repeat CXR was performed- a few solid dense findings in the right lung.
            Therefore, she was referred for a PET-CT which demonstrated increased uptake in the right
            breast, lymph nodes on the right a few areas in the lungs and liver.
            On biopsy from the lesion in the right breast- triple negative adenocarcinoma. Genetic
            testing has not been done thus far.
            Genetic counseling- the patient denies a family history of breast, ovary, uterus,
            and prostate cancer. Her mother has chronic lymphocytic leukemia (CLL).
            She is planned to undergo genetic tests because the aggressive course of the disease,
            and her young age.
            Impression:
            Stage 4 triple negative breast adenocarcinoma.
            Could benefit from biological therapy.
            Different treatment options were explained- the patient wants to get a second opinion."""

patient_document2 = models.PatientDocument(type=models.DocumentType.NOTE,
                                           id="doc2",
                                           content=models.DocumentContent(
                                               source_type=models.DocumentContentSourceType.INLINE,
                                               value=doc_content2),
                                           clinical_type=models.ClinicalDocumentType.PATHOLOGY,
                                           language="en",
                                           created_date_time=datetime.datetime(2021, 10, 20))

doc_content3 = """
            PATHOLOGY REPORT
                                    Clinical Information
            Ultrasound-guided biopsy; A. 18 mm mass; most likely diagnosis based on imaging:  IDC
                                        Diagnosis
            A.  BREAST, LEFT AT 2:00 4 CM FN; ULTRASOUND-GUIDED NEEDLE CORE BIOPSIES:
            - Invasive carcinoma of no special type (invasive ductal carcinoma), grade 1
            Nottingham histologic grade:  1/3 (tubules 2; nuclear grade 2; mitotic rate 1;
            total score; 5/9)
            Fragments involved by invasive carcinoma:  2
            Largest measurement of invasive carcinoma on a single fragment:  7 mm
            Ductal carcinoma in situ (DCIS):  Present
            Architectural pattern:  Cribriform
            Nuclear grade:  2-
                            -intermediate
            Necrosis:  Not identified
            Fragments involved by DCIS:  1
            Largest measurement of DCIS on a single fragment:  Span 2 mm
            Microcalcifications:  Present in benign breast tissue and invasive carcinoma
            Blocks with invasive carcinoma:  A1
            Special studies: Pending"""

patient_document3 = models.PatientDocument(type=models.DocumentType.NOTE,
                                           id="doc3",
                                           content=models.DocumentContent(
                                               source_type=models.DocumentContentSourceType.INLINE,
                                               value=doc_content3),
                                           clinical_type=models.ClinicalDocumentType.PATHOLOGY,
                                           language="en",
                                           created_date_time=datetime.datetime(2022, 1, 1))

patient_doc_list = [patient_document1, patient_document2, patient_document3]
patient1.data = patient_doc_list
# <\DocumentList>

# Set configuration to include evidence for the cancer staging inferences
configuration = models.OncoPhenotypeModelConfiguration(include_evidence=True)

# Construct the request with the patient and configuration
cancer_profiling_data = models.OncoPhenotypeData(patients=[patient1], configuration=configuration)

poller = await cancer_profiling_client.begin_infer_cancer_profile(cancer_profiling_data)
cancer_profiling_result = await poller.result()
if cancer_profiling_result.status == models.JobStatus.SUCCEEDED:
    results = cancer_profiling_result.results
    for patient_result in results.patients:
        print(f"\n==== Inferences of Patient {patient_result.id} ====")
        for inference in patient_result.inferences:
            print(
                f"\n=== Clinical Type: {str(inference.type)} Value: {inference.value}\
                    ConfidenceScore: {inference.confidence_score} ===")
            for evidence in inference.evidence:
                data_evidence = evidence.patient_data_evidence
                print(
                    f"Evidence {data_evidence.id} {data_evidence.offset} {data_evidence.length}\
                        {data_evidence.text}")
else:
    errors = cancer_profiling_result.errors
    if errors is not None:
        for error in errors:
            print(f"{error.code} : {error.message}")

Problembehandlung

Allgemein

Die Health Insights Cancer Profiling-Clientbibliothek löst ausnahmen aus, die in Azure Core definiert sind.

Protokollierung

Diese Bibliothek verwendet die Standardprotokollbibliothek für die Protokollierung.

Grundlegende Informationen zu HTTP-Sitzungen (URLs, Header usw.) werden auf ebener Ebene INFO protokolliert.

Die Protokollierung auf detaillierter DEBUG Ebene, einschließlich Anforderungs-/Antworttexten und nicht ausgeführten Headern, kann auf dem Client oder pro Vorgang mit dem logging_enable argument Schlüsselwort (keyword) aktiviert werden.

Die vollständige SDK-Protokollierungsdokumentation mit Beispielen finden Sie hier.

Optionale Konfiguration

Optionale Schlüsselwort (keyword) Argumente können auf Client- und Vorgangsebene übergeben werden. In der Azure Core-Referenzdokumentation werden verfügbare Konfigurationen für Wiederholungen, Protokollierung, Transportprotokolle und vieles mehr beschrieben.

Nächste Schritte

Zusätzliche Dokumentation

Eine ausführlichere Dokumentation zur Krebsprofilerstellung in Azure Health Insights finden Sie in der Dokumentation zur Krebsprofilerstellung auf docs.microsoft.com.

Mitwirken

Beiträge und Vorschläge für dieses Projekt sind willkommen. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. Ausführliche Informationen finden Sie unter https://cla.microsoft.com.

Wenn Sie einen Pull Request (PR) übermitteln, überprüft ein CLA-Bot automatisch, ob Sie eine Lizenzvereinbarung bereitstellen und den PR entsprechend ergänzen müssen (z.B. mit einer Bezeichnung oder einem Kommentar). Führen Sie einfach die Anweisungen des Bots aus. Sie müssen dies nur einmal für alle Repositorys ausführen, die unsere CLA verwenden.

Für dieses Projekt gelten die Microsoft-Verhaltensregeln für Open Source (Microsoft Open Source Code of Conduct). Weitere Informationen finden Sie in den FAQ zum Verhaltenskodex, oder wenden Sie sich mit weiteren Fragen oder Kommentaren an opencode@microsoft.com.