Related Experiment Videos
Medical image processing utilizing neural networks trained on a massively parallel computer
1Biomedical Engineering Program, Iowa State University, Ames, USA.
Computers in Biology and Medicine
|July 1, 1995
Summary
Large artificial neural networks (ANNs) can now be trained for medical image processing using parallel computing. This enables faster processing of novel images on standard computers after initial training.
Area of Science:
- Medical Imaging
- Computer Science
- Artificial Intelligence
Background:
- Artificial neural networks (ANNs) are typically limited to small architectures.
- Medical image processing applications often require large neural network architectures.
Purpose of the Study:
- To demonstrate the feasibility of training very large architecture ANNs for medical image processing.
- To leverage massively parallel computing for accelerating ANN training.
Main Methods:
- Utilized a massively parallel, single-instruction multiple data (SIMD) computer for training.
- Trained several ANNs for tomographic reconstruction of 64x64 single photon emission computed tomography (SPECT) images from 64 planar views.
Main Results:
- Achieved a two- to three-orders of magnitude improvement in processing time.
- Demonstrated practical training of very large architecture ANNs.
Conclusions:
- Large architecture ANNs can be effectively trained for medical image processing using parallel computing.
- Trained neural networks can be deployed on serial computers (PCs, workstations) for efficient processing of novel images.