Why

The Client is a multinational corporation that develops medical devices and pharmaceuticals. They produce robotic devices for performing surgery. The Client needs an advanced analytics platform to collect, analyse and integrate data from robotic devices in multiple locations. The aim is to allow hospital staff to run applications based on a unified source of information.

What & how

Datumo built the data platform from scratch which supports multiple use-cases and is easily scalable. It abides by strict medtech industry compliance regulations. ‍

Azure Cloud is used. The platform is split into 5 main areas:

• device data ingestion and integration
• hospital data integration and analytics
• Advanced Analytics use cases, such as:
  - monitoring the number of operations performed by a device and notifying the support team that it needs to be replaced after the specified threshold is exceeded
  - flexible and ad hoc analytics, conducted globally in minutes instead of hours by many people simultaneously orchestration and operational support of the application
• MLOps pipelines to run ML models in production and support experiments.

The platform analyses the use of robotic devices during surgeries. Thanks to this, the tools can better support hospital staff. Device data is gathered and securely stored in a partitioned manner and then transformed into queryable SQL format. The platform also allows operators to send commands to devices such as configuration changes.

Hospital data integration, based on API management, allows the gathering of patient data available in medical records. Data is correctly mapped to particular patients. Having centralised medical records makes it easier for doctors and nurses to find relevant information and make more efficient decisions.

The Advanced Analytics system is designed to support Big Data use cases. It includes the partitioning of data and custom ETL processing. It maximised efficiency in data governance, introduced better data organisation and streamlined maintenance.

Orchestration and operational support is focused on onboarding customised user applications. It simplified the management of applications and lowered the costs of the support, while increasing the SLA and quality through visualisation of codebase that can be written in multiple programming languages in order to monitor and debug solutions.

MLOps pipelines are created to support Data Scientists who prepare models and run experiments on datasets provided by devices. As a result ML model deployment is much faster and the accuracy of delivered models is substantially higher. This was achieved through integration of the Advanced Analytics component and usage of MLFlow which allows for faster iterations, cooperation with Data Engineers, detailed monitoring and tuning of ML experiments.

Datumo was responsible for:

• architecting end to end solution
• deployment and qualification of the Airflow solutions running on Kubernetes engine
• Dockerization of the applications and onboarding them to Airflow
• deployment and qualification of the Databricks platform used to run ETL pipelines
• customization of MLFlow pipelines to support MLOps
• consulting the Azure ADLS use cases and adjusting partitioning options
• R&D development and assessment of the Azure based technologies such as Cosmos DB, EventHub, EventGrid
• integration with the Databricks AutoLoader feature to support real-time and streaming use-cases of device data integration from ADLS directly into Delta Tables
• presenting the solution and performing knowledge transfer sessions for users and business stakeholders
• management of the development teams responsible for Advanced Analytics.

Benefits

• support for many cases allowing scalability, cost optimisation, as well as an increase in quality and SLA
• adherence to strict compliance regulations of the medtech industry
• analysis of the usage of robotic devices during surgeries allowing the tools to better support hospital staff
• centralisation of patient data in one place resulting in more efficient decision-making
• faster iterations, detailed monitoring and tuning of ML experiments thanks to more accurate ML models
• sharing best practices -presenting the solution and performing knowledge transfer sessions for tenants and business stakeholders.