Deep Learning in Medical Post-Processing

The challenge

A well-known manufacturer of medical image post-processing applications and long-time UL Solutions Software Intensive Systems (SIS) customer asked SIS engineers to support its evaluation of new technology trends in the algorithmic image processing domain. The customer aimed to improve product quality while maintaining the same level of product suitability for daily use.

The primary goal was the evaluation of opportunities in machine learning — particularly deep learning — and, if applicable, their integration into existing medical devices. SIS supported this customer on its journey from the prototype to the integration of robust deep learning-based algorithms.

Project duration and resourcing 

• About 1.5 years
• Two .net and C++ developers

Our approach

First, an established open-source deep learning library was introduced into the international multi-team project setup. The library was adapted to meet the project’s technological requirements and licensing rights, with consideration of the customer’s existing processes and relevant norms and standards in the regulated medical environment.

Principles of software craftsmanship were considered and implemented consistent with industry standards. Additionally, all tasks were mapped to numerous requirements of the medical and regulatory environment, including the creation and maintenance of comprehensive technical documentation.

SIS evaluated the customer's compliance with requirements in terms of code efficiency, execution times and consumption of software resources on customer-specific hardware configurations. In the course of development, SIS worked closely with clinical experts. SIS engineers were contact persons for all questions from the customer’s technical experts who used the new implemented functionalities. Additionally, SIS was in constant communication with medical algorithmic image processing scientists.

SIS engineers integrated the new machine learning technology into the existing comprehensive code base of the customer project. Our work included preventive measures against unwanted side effects on algorithmic results due to changes in central components.

Highlights of our work

• Introduction of an open-source deep learning library
• Consideration of software craftsmanship rules
• Integration of latest clinical and scientific knowledge

Customer benefits

With the support of SIS, the customer successfully implemented the new deep learning technology, which helped the customer improve medical imaging software quality and strengthen its position as a technological leader.

Key benefits to our customer included:

• Early application of the technological trend deep learning.
• Integration of new technologies consistent with industry standards.
• Improved software quality.