Skip to main content

Software Development using Artificial Intelligence

Through the technology of Artificial Intelligence (AI) image segmentation, contouring, object detection, classification and regression tasks are now available with unseen accuracy and computational speed in medical workflows. We offer this technology in a customer-centered workflow and help you to reduce the time to market. We provide customer-based services that combine all stages necessary to incorporate AI into your applications. This comprises data annotation, network design, network training on dedicated hardware, adaptations of model design, and integration into your existing software to achieve outstanding results.

Our vision is to bring AI solutions and our expertise into your business.

 

Stages of our AI Services

Data Annotation

  • In-house efficient annotation tools for 2D, 3D and 4D data 
  • Annotation by our medical experts
  • Defined quality standards for annotations
  • Planning and support with data acquisition
  • Build customized anatomical data bases

Benefits

  • Experience from successful past projects
  • Receive annotations in standard formats
  • Keep ownership of data and annotations

Model Design and Training

  • Custom neural network design

  • Problem-specific pre-processing

  • Dedicated high performance hardware

Benefits

  • Learn from our experienced AI experts

  • Reduce engineering time

  • Fast evolving technology

Customer Integration

  • Delivery of trained model and annotated data
  • Software SDK and API (C / C++)
  • Source code of model design
  • API for hassle-free integration

Benefits

  • Certified expert medical software development
  • Model reusable for future applications
  • Platform-independent software
  • CPU and GPU supported

Lifetime Management and Support

  • Refinement and adaptions during clinical use-test
  • Re-training with additional data
  • Keep model design and software up to date

Benefits

  • State-of-the-art technology
  • Experience in regulatory requirements

Features of our AI Services

  • Multi-modal input data

  • Single-click solutions

  • No parameterizations necessary

  • API for hassle-free integration

  • Complete Package from Annotation of the Data to Integration

Work Cases using AI

Automated Organ Segmentation on MRI

In this example we realized a robust and accurate segmentation of the liver, the spleen, the kidneys and the heart in various Magnetic Resonance Imaging (MRI) sequences. The apporach has proven to cope with even non-standardized image intensities and varying contrast in MRI. All aspects of a typical AI service project have been realized in-house by our annotation and software experts. We have achieved an average DICE coefficient across the sequences of 92%.

Image Denoising with Deep Learning

In this example we realized the denoising of X-Ray images of the spine. Deep Learning techniques allow to predict the noise of an image. The predicted noise thus can be removed from an image which results in a denoised image.

Chimaera Model Compression Technology

Medical applications in computer aided diagnosis like fluoroscopy require real-time capable algorithms which can process high frame rates. In the same time there is a growing demand for image processing applications on low-end hardware like mobile devices or embedded solutions. We have developed an algorithm that in a smart way reduces the number of parameters with almost no effect on the accuracy.

We have trained and tested our algorithms with a dataset of 67 images of a seven-class segmentation problem. The target organs are: right and left lung, liver, spleen, right and left kidney. This data set was also used for a two-class-segmentation problem (i.e. differentiating tissue from background). As basis for our compressed algorithm we used the SegNetmodel.

 Number of ParametersAccuracy

Reduction of Parameters in %

SegNet7-class problem

29,458,2110.930-

Our Solution with reduced Parameters

SegNet 7-class problem

7,881,722

 

0.927

 

73.2 %

 

SegNet  2-class problem

29,457,5520.940

-

Our Solution with reduced Parameters

SegNet 2-class problem

1,213,3250.929

95.9 %

With our algorithm we improved the frame rate by the factor 3.5 and reduced the memory usage by more than 60%. This applications enables new possibilities especially for mobile devices and real-time solutions.

Vijay Badrinarayanan, Alex Kendall, and Roberto Cipolla. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. CoRR, abs/1511.00561, 2015.

Download Material

  price
  demo
  Contact