MLonFHIR: Fusing Sklearn with the HL7 FHIR Standard

栏目: IT技术 · 发布时间: 4年前

内容简介:Read more about internals and extendability in theThere are two general ways of searching for patients with specific properties. The first one is to search by coding system:The second one is by text. The searched tags are CodeableConcept.text, Coding.displ

ml-on-fhir

A work in progress library that fuses the HL7 FHIR standard with scikit-learn

Read more about internals and extendability in the readthedocs .

Usage (taken from our demo notebook )

First: Register the base URL of your database with a FHIRCient object:

from fhir_client import FHIRClient
import logging
import pandas as pd

logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)

client = FHIRClient(service_base_url='https://r3.smarthealthit.org', logger=logger)

Querying Patients

There are two general ways of searching for patients with specific properties. The first one is to search by coding system:

# To receive a list of available procedures:
procedures = client.get_all_procedures()
pd.DataFrame([prod.code['coding'][0] for prod in procedures]).drop_duplicates().sort_values(by=['display']).head()

# Now retrieve patients
patients_by_procedure_code = client.get_patients_by_procedure_code("http://snomed.info/sct","73761001")

The second one is by text. The searched tags are CodeableConcept.text, Coding.display, or Identifier.type.text:

conditions = client.get_all_conditions()
pd.DataFrame([cond.code['coding'][0] for cond in conditions]).drop_duplicates(subset=['display']).sort_values(by='display', ascending=True).head()

patients_by_condition_text = client.get_patients_by_condition_text("Abdominal pain")

One can also load a control group for a specific cohort of patients. The control group is of equal size of the case cohort (min size: 10) and is composed of randomly sampled patients that do not match the original query. Their class is contained in the .case property of the Patient object.

patients_by_condition_text_with_controls = client.get_patients_by_condition_text("Abdominal pain", controls=True)

print("{} are cases and {} are controls".format(len([d for d in patients_by_condition_text_with_controls if d.case]), 
                                                len([d for d in patients_by_condition_text_with_controls if not d.case])))

Machine Learning

To train a classifier, we need to first tell the MLOnFHIRClassifier the type of object which we would like to classify. We can then define features ( feature_attrs ) and labels ( label_attrs ) for our classification task and pass the preprocessor of our current client, so it is clear how to preprocess the features/labels of a patient. We can then simply call .fit on the MLOnFHIRClassifier instance together with our classifier of choice.

from ml_on_fhir import MLOnFHIRClassifier
from fhir_objects.patient import Patient
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score, roc_curve, auc

ml_fhir = MLOnFHIRClassifier(Patient, feature_attrs=['birthDate', 'gender'],
                             label_attrs=['case'], preprocessor=client.preprocessor)
X, y, trained_clf = ml_fhir.fit(patients_by_condition_text_with_controls, DecisionTreeClassifier())

from sklearn.metrics import accuracy_score, roc_curve, auc
fpr, tpr, _ = roc_curve(y, trained_clf.predict(X))
print("Prediction accuracy {}".format( auc(fpr, tpr) ) )

以上就是本文的全部内容,希望本文的内容对大家的学习或者工作能带来一定的帮助,也希望大家多多支持 码农网

查看所有标签

猜你喜欢:

本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们

The Ruby Programming Language

The Ruby Programming Language

David Flanagan、Yukihiro Matsumoto / O'Reilly Media, Inc. / 2008 / USD 39.99

Ruby has gained some attention through the popular Ruby on Rails web development framework, but the language alone is worthy of more consideration -- a lot more. This book offers a definition explanat......一起来看看 《The Ruby Programming Language》 这本书的介绍吧!

MD5 加密
MD5 加密

MD5 加密工具

HEX CMYK 转换工具
HEX CMYK 转换工具

HEX CMYK 互转工具