Deep Learning Inference: An application to screening lung diseases.
Principal Investigator
Name
Wakahiu Njenga
Degrees
BS. EECS, MS. CE
Institution
http://behold.ai/
Position Title
Chief Technical Officer
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-235
Initial CDAS Request Approval
Aug 18, 2016
Title
Deep Learning Inference: An application to screening lung diseases.
Summary
The advent of deep learning has revolutionized the field of computer vision and machine perception. Deep learning technology combines artificial neural network architectures with massive computing power to perform training and inferences. We would like to perform highly complex pattern recognition and highlight conspicuous abnormalities in medical scans and images such as lungs nodules. Our goal is to test the efficacy of deep learning algorithms and to bench-mark them against existing screening and CADx technologies.
Project steps:
Step 1: label data
Prepare lung cancer images, benign nodule patients, and other lung images.
Step 2: Data training
Train labeled lung cancer medical image. 60% of the data will be used as training data set. Training involves tuning a neural network's numeric weights based on experience - i.e. determining its parameters using labeled examples of inputs and desired output. This makes neural nets adaptive to inputs and capable of learning.
Step 3: Evaluation of algorithms
We will deploy our network to run inferences i.e. using its previously trained parameters to classify, recognize, and generally process unknown but similar inputs. 20% of the dataset will be used as validation set and the last 20% will be used as test data set.
Details about CAD algorithms
Purpose:
The purpose is to detect lung cancer automatically using deep learning from benign nodules or other lung diseases, such as ILD.
Methods:
We are planning to use deep convolutional neural networks (CNN) and deep contract learning method to train lung medical images.
Project steps:
Step 1: label data
Prepare lung cancer images, benign nodule patients, and other lung images.
Step 2: Data training
Train labeled lung cancer medical image. 60% of the data will be used as training data set. Training involves tuning a neural network's numeric weights based on experience - i.e. determining its parameters using labeled examples of inputs and desired output. This makes neural nets adaptive to inputs and capable of learning.
Step 3: Evaluation of algorithms
We will deploy our network to run inferences i.e. using its previously trained parameters to classify, recognize, and generally process unknown but similar inputs. 20% of the dataset will be used as validation set and the last 20% will be used as test data set.
Details about CAD algorithms
Purpose:
The purpose is to detect lung cancer automatically using deep learning from benign nodules or other lung diseases, such as ILD.
Methods:
We are planning to use deep convolutional neural networks (CNN) and deep contract learning method to train lung medical images.
Aims
Our goal is to test the efficacy of deep learning algorithms and to bench-mark them against existing screening and CADx technologies.
Collaborators
Jing Chao Lin (Behold.ai, MIT) jing@ behold.ai