Comparing accuracy of prevalent Neural Network models for classification of Radiology images
Principal Investigator
Name
Keyur Mithawala
Degrees
MBA, B.Engg.
Institution
None
Position Title
Researcher - Neural Networks and Deep Learning
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-226
Initial CDAS Request Approval
Jul 5, 2016
Title
Comparing accuracy of prevalent Neural Network models for classification of Radiology images
Summary
I am researching multiple Neural Network models including Regional, Convolution, and ResNet with a focus on their potential applications in Radiology image classification.
All current evaluations of these models use ImageNet dataset, which includes huge variation in low-resolution images, RGB color, and 10,000 categories across the dataset. Therefore, ImageNet dataset is not a good indicator of performance of these NN models for medical imaging classification with limited variation, no color, higher resolution, and 300+ categories.
I would like to obtain large Radiology imaging dataset corpora and use it for evaluating classification performance of NN models in a domain-specific application for Radiology.
All current evaluations of these models use ImageNet dataset, which includes huge variation in low-resolution images, RGB color, and 10,000 categories across the dataset. Therefore, ImageNet dataset is not a good indicator of performance of these NN models for medical imaging classification with limited variation, no color, higher resolution, and 300+ categories.
I would like to obtain large Radiology imaging dataset corpora and use it for evaluating classification performance of NN models in a domain-specific application for Radiology.
Aims
Rank healthcare-specific classification accuracy results of prevalent NN models like CNN, ResNet, RCNN, LeNet, etc. against Radiology Imaging dataset.
Collaborators
None