Azure认知服务健康洞察癌症分析Python客户端库

健康洞察 是一个使用Azure认知服务框架构建的Azure应用人工智能服务，利用多个认知服务、医疗API服务和其它Azure资源。

癌症分析模型 接收肿瘤科患者的临床记录，并输出癌症分期，如临床分期TNM类别和病理分期TNM类别以及肿瘤位置、组织学。

源代码 | 包（PyPI） | API参考文档 | 产品文档 | 示例

入门指南

先决条件

• 使用此包需要Python 3.7或更高版本。
• 您需要一个Azure订阅 来使用此包。
• 现有的认知服务健康洞察实例。

安装包

pip install azure-healthinsights-cancerprofiling

此表显示了SDK版本与服务支持的API版本之间的关系

验证客户端

获取端点

您可以使用Azure门户或Azure CLI找到您的健康洞察服务资源的端点

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

获取API密钥

您可以从Azure门户中的健康洞察服务资源获取API密钥。或者，您可以使用以下Azure CLI片段获取资源的API密钥。

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

使用API密钥凭据创建CancerProfilingClient

一旦您获得了API密钥的值，您就可以将其作为字符串传递给一个AzureKeyCredential实例。使用该密钥作为凭据参数来验证客户端

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))

长时间运行的操作

长时间运行的操作是指由向服务发送的初始请求来启动操作的操作，然后以一定的时间间隔轮询服务以确定操作是否完成或失败，如果成功，则获取结果的操作。

支持医疗分析、自定义文本分析或多个分析的方法被建模为长时间运行的操作。客户端公开一个返回轮询对象的begin_<method-name>方法。调用者应该通过在从begin_<method-name>方法返回的轮询对象上调用result()来等待操作完成。以下提供了示例代码片段来演示如何使用长时间运行的操作（见下文）。

关键概念

Cancer Profiling模型允许您从非结构化的临床文档中推断癌症属性，如肿瘤部位、组织学、临床阶段TNM类别和病理阶段TNM类别。

示例

以下部分提供了几个代码片段，涵盖了健康洞察 - 癌症分析服务的一些最常见任务，包括

• 癌症分析

癌症分析

从患者的非结构化临床文档中推断关键癌症属性，如肿瘤部位、组织学、临床阶段TNM类别和病理阶段TNM类别。

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}")

故障排除

一般

健康洞察癌症分析客户端库将引发定义在Azure Core中的异常。

日志记录

此库使用标准的logging库进行日志记录。

HTTP会话的基本信息（URL、头等）以INFO级别记录。

详细的DEBUG级别日志记录，包括请求数据和响应体以及未删除的头，可以通过在客户端或每个操作中使用logging_enable关键字参数来启用。

有关示例的完整SDK日志记录文档，请参阅此处。

可选配置

可以在客户端和每个操作级别传递可选关键字参数。azure-core 参考文档描述了重试、日志记录、传输协议等可用配置。

下一步

其他文档

有关Azure健康洞察癌症分析的更详细文档，请参阅docs.microsoft.com上的癌症分析文档。

贡献

本项目欢迎贡献和建议。大多数贡献需要您同意贡献者许可协议（CLA），声明您有权并且确实授予我们使用您的贡献的权利。有关详细信息，请访问 https://cla.microsoft.com。

当您提交拉取请求时，CLA-机器人会自动确定您是否需要提供CLA，并相应地装饰PR（例如，标签，注释）。只需按照机器人提供的说明操作。您只需在整个使用我们的CLA的仓库中这样做一次。

本项目采用了 微软开源行为准则。有关更多信息，请参阅行为准则FAQ或通过 opencode@microsoft.com 联系我们，提出任何额外的问题或评论。