Distributed Automated Tuberculosis and Cancer Diagnosis
Principal Investigator
Name
Jochen Nessel
Degrees
Ph.D.
Institution
Rai and Rohl Technologies, Inc.
Position Title
Director - Machine Learning
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-256
Initial CDAS Request Approval
Feb 22, 2017
Title
Distributed Automated Tuberculosis and Cancer Diagnosis
Summary
There is a lot of work around using single algorithms on specific data sets and evaluate performance. We aim to use several classifiers and combine them using different versions of Boosting to increase performance of the so built "strong" classifier.
To achieve a reasonable level of performance, the "weak" classifiers should work in a distributed environment.
The infrastructure should allow for weak classifiers to be updated without having to stop the overall strong classifier. Also, it should be possible to update the strong classifier once one or more weak classifiers have been updated.
We would like to use as many data sources and as much data as we can get to build a system that will be able to scale when
- new data becomes available,
- new weak classifiers are available,
- and there are potentially a lot of images run through the system.
We intend to use a number of ML algorithms as the weak classifiers, including, but not limited to, Neural Networks ("deep learning"), SVM, decision trees, Bayes-based approaches, and others.
To achieve a reasonable level of performance, the "weak" classifiers should work in a distributed environment.
The infrastructure should allow for weak classifiers to be updated without having to stop the overall strong classifier. Also, it should be possible to update the strong classifier once one or more weak classifiers have been updated.
We would like to use as many data sources and as much data as we can get to build a system that will be able to scale when
- new data becomes available,
- new weak classifiers are available,
- and there are potentially a lot of images run through the system.
We intend to use a number of ML algorithms as the weak classifiers, including, but not limited to, Neural Networks ("deep learning"), SVM, decision trees, Bayes-based approaches, and others.
Aims
- Distributed, scalable diagnosis system
- Incorporate data from many sources
Collaborators
Jochen Nessel, Rai and Rohl Technologies, Inc.
Cuong "Ken" Nguyen, Rai and Rohl Technologies, Inc.
Cory Gordon Sherman, Rai and Rohl Technologies, Inc.