Connecting low code / no code tools to Machine Learning Models is often discussed but rarely implemented in the real world. The video above demonstrates near-real-time predictions in a Power App using Azure ML to make a presumptive medical diagnosis based on answers about symptoms entered by a user. Symptom data entered into a Power App within Microsoft Teams is passed to Azure ML via Power Automate, scored by the Machine Learning Model, logged in the Dataverse, and then seamlessly returned to the Power App in Teams.
The demo predicts Acute Inflammation of Urinary Bladder (Cystitis) and/or Acute Nephritis of Renal Pelvis, which are separate diagnoses often having overlapping symptoms. The University of California Irvine provides an Open Source dataset cataloguing assessments of patients presenting with symptoms related to these diagnoses, which was used to train the Machine Learning Model. Links to the source data and reference articles are included in the video description at this link. This Demo is not intended to be used in the real world (even though the data is real), but is for the purpose of demonstrating the ease by which these types of Presumptive Diagnostic Algorithms can be deployed to achieve Realized Value. Once a machine learning model has been built in Azure ML, the front end interface can be built out in a few hours using low code / no code Microsoft tools.
The video also walks through the architecture for ingesting the ML training data using Azure Data Factory, Azure Data Lake, analyzing the training data with Power BI, and then building an Azure ML Predictive Model. Presumptive Diagnostics are a complicated topic requiring due diligence by medical professionals, but when implemented ethically and with scientific rigor can provide benefits to quality and availability of care.
Everything in this demo is 100% real and operational. As I'm writing this article, it is available in the Microsoft Healthcare Demo environment. If you are a Microsoft Customer, you can ask your field representative to give you a tour of this solution.
