AI is currently the new big thing and it’s really here, this is based on an FDA approved platform- TeraRecon.
1.What is it?
An AI tool which accurately identifies cancerous cells from healthy cells using an image- recognition algorithm with extreme precision boundaries.
The system uses a Natural User Interface (NUI) which is compatible with PACS via TeraRecon.
2. Why is it important?
This can be used across multiple specialties such as:
- radiology
- oncology
- surgery
Essentially saving time and cost whilst enhancing medical diagnosis skills to increase time to optimise patient treatment plans.
Also with regards to patient data safety, all images are anonymised and encrypted before entering the programme, reducing concerns for hacking results.
Results show similar and higher accuracy compared to baseline architectures along with reduced computational cost, producing results in better time, for example with ResNet 50 illustrates:
- 31% faster on a CPU
- 40% fewer parameters
3. How does it work?
- CT and MRI images are sent to Project InnerEye for automatic delineation
- The system completes a pixel by pixel analysis to contour organs around a tumour site using Deep Decision Forests (DDFs) and Convolution Neural Networks (CNNs) for 3D voxel analysis
- A colour analysis is performed, checking for colour, brightness and intensity, to identify the boundaries between healthy and non- healthy cells
- The boundaries can then be used for quantitative radiology and more efficient planning of radiotherapy and surgery
*A voxel is a 3D pixel
4. The future
Image analytics algorithms by Project InnerEye will be integrated in third party systems enable advances in:
- Haematology imaging – malaria diagnosis
- 2D xrays
- Higher dimensional images
- International accessibility to allow a more fluid treatment process across different clinical settings safely
5. Who is involved?
PI – Antonio Criminisi, Microsoft Research, Cambridge UK
EnvoyAI
Collaborations with:
- University of Oxford
- University of Cambridge
- John Hopkins Medical Institute
- Cornell Medical School
- Massachusetts General Hospital
- University of Washington
- Kings College London
References
Ioannou Y, Robertson D, Cipolla R, Criminisi A. Deep Roots: Improving CNN Efficiency with Hierarchical Filter Groups. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017.
https://www.microsoft.com/en-us/research/wp-content/uploads/2016/06/ICCV15_DeepNDF_main.pdf
https://www.envoyai.com/blog/terarecon-demonstrates-project-innereye-radiomics-powered-by-envoyai
https://www.microsoft.com/en-us/research/project/medical-image-analysis/
One Comment Add yours