AI/ML Case Study

Device Tracking Improvements with Azure Machine Learning Implementation for a Global Med Device Company

Client

The client is a global medical technology company specializing in the development of innovative solutions for treating cardiovascular diseases and neuromodulation disorders. Their product portfolio focuses on improving patient outcomes in areas such as cardiac surgery, neuromodulation for epilepsy, and sleep apnea therapy. With operations across more than 100 countries, the company aims to advance healthcare through cutting-edge technology and research.

CLIENT PROFILE
EMPLOYEES: 20,000+
INDUSTRY: Medical Devices / Manufacturing
FOUNDED: 1987

Project Scope

Current tracking processes are not as thorough

Since the company is a global company, they need to ship devices all over the world. Each device needs to be tracked for various reasons including compliance, however, with their current system, they are not able to completely and thoroughly track the devices as well as they could especially when there are anomalies. The shipping data comes from disparate sources, from different systems and at different points in time. The client would like to bring them all together to create one combined shipping database that can then be used in reports, dashboards, etc.

Smartbridge has extensive experience in Azure implementations

The client needed to migrate their data and processes to the cloud, so they can scale their operations, securely access the data they needed when they needed it and improve existing pipelines and establish new ones. Therefore, the main objective for them is to move their data science workloads to the Azure Machine Learning (Azure ML) platform. Azure ML addresses traditional machine learning challenges by offering automated processes, robust data security, and scalable cloud infrastructure.

Key Challenges

A local system and manual handling of data creates hurdles

  • Limited integration of MLOps practices
  • Data was downloaded from the source to local computer with public internet access, increasing the risk of potential leakage
  • Data retrieval through R code from various sources. Data access restricted to data scientists, requiring approval
  • Absence of data science assets version control, Manual management of large code volumes
  • Local compute used exclusively
  • Development silos in the local environment which makes the hand-off process difficult
  • Manual scheduling for script execution and Lack of monitoring during data processing
  • Anomaly detection ML effort in planning phase

The Smartbridge Solution

Implementing an enterprise-level Azure Machine Learning program

The Smartbridge team used MLOps (Machine Learning Operations) practices to help the client enable collaboration and streamline their processes. MLOps is a set of practices that combine machine learning, DevOps, and data engineering principles to automate and standardize the development, deployment, and monitoring of machine learning models. By implementing MLOps, the team ensured that the machine learning lifecycle—covering model development, testing, deployment, and ongoing monitoring—was fully integrated into a scalable and efficient process.

MLOps

The team started by transitioning disparate data science workloads to the Azure machine learning platform, which not only enabled seamless integration, scalability, and security but also provided access to a wide range of tools and services necessary for efficient model management.

In addition to migrating to Azure ML, the team re-factored the project to align with the ML lifecycle, making the solution monitorable, maintainable, and adaptable to future changes. By building a pipeline and organizing each step as an individual block, the team ensured that the system is highly modular. This modularity offers several benefits:

  • Reusability of blocks across different projects, reducing duplication of effort.
  • Different computing resources can be allocated based on the complexity of each block, optimizing resource usage. For instance, resource-heavy models can use powerful instances, while simpler models can run on lighter infrastructure.
  • Continuous monitoring and maintenance of the solution become easier because individual blocks can be updated or modified without impacting the entire workflow, allowing for smooth model updates and faster iterations.

When using Azure, Microsoft recommends using Python, but as previously stated, the client’s data science team uses R as their programming language. Smartbridge solved this discrepancy by creating a Python wrapper to go around the R code, so the client can still use codes they previously created and feel comfortable continuing using the language they are used to until they can pick up Python.

To resolve the issue of security, the team put the data in a secured storage container and create a secured connection to the Azure ML environment. This eliminates the need for data to be downloaded to local machines in order to be used allowing the company to stay in full compliance with corporate and healthcare policies.

Finally, Smartbridge was also able to implement versioning for models which allowed the client to retrain models without losing their original versions, so they can do things like compare and validate results.

Overall Solution:

  • Data science team is trained on MLOps principles and following MLOps best practices through Azure ML
  • Data only moved within closed VNet for the end-to-end data science development, eliminating any chance of leakage
  • Raw data for data science team all land in one secure location in Azure before being accessed by the data scientists
  • Comprehensive version control for data assets, models, pipelines in Azure ML, and code version control via Azure DevOps integration
  • Flexible compute options: local or cloud-based
  • Centralized repository for development code and data science assets, facilitating collaboration and project transitions
  • Automated pipeline scheduling and data logging for data preparation, ML training, and inferencing, minimizing human intervention
  • Quickly developed anomaly detection ML with simplified processes through Azure ML.

Prior to the Smartbridge solution, the data science team was restricted to sending the data to its stakeholders and letting them figure out the anomalies. In addition, identifying an anomaly in the shipping process and finding who to talk to about it was a manual, repetitive process. Now with all of the data aggregated in one place with valuable insights, business users are able to access the information they need without having to waste time tracking it down.

Value Delivered

Creating a 87% reduction in processing time

Without Azure machine learning, the anomaly detection process could take nearly an entire workday. With an automation pipeline in Azure ML, it would take less than an hour.

Keep Reading: Forecasting Improvements with Azure Machine Learning Implementation for a Global Medical Device Company

Share This:

There’s more to explore at Smartbridge.com!

Sign up to be notified when we publish articles, news, videos and more!

By signing up for emails from Smartbridge.com, you agree to our terms and privacy policy.