Big Data Assisted Lung Nodule Detection
Principal Investigator
Name
Timo Bremer
Degrees
Ph.D.
Institution
Lawrence Livermore National Laboratory
Position Title
Research Scientist/Project Leader
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-120
Initial CDAS Request Approval
Feb 13, 2015
Title
Big Data Assisted Lung Nodule Detection
Summary
Automatically processing and classifying radiologic exams has the
potential to drastically lower the costs of large scale screenings
while improving patient outcomes especially in areas such as lung or
breast cancer. While a number of attempts using various image
processing or computer vision techniques have been proposed over the
last decade the current state of the art remains restricted to
specific instances and produces unremarkable results. One particular
challenge is that the collected data is often noisy and contains significant
patient to patient variations.This makes it difficult to train
traditional approaches as collecting labeled data that covers all
possible variation is virtually impossible. Instead, recent advances
in big data related technique promise similar or better results
without the need for labeled training data. More specifically, given
enough data such systems can exploit statistics of large numbers to
reliably detect objects in images. This data set will be of
substantial value to further develop, test, and evaluate such
techniques as a precursor to more wide scale studies.
potential to drastically lower the costs of large scale screenings
while improving patient outcomes especially in areas such as lung or
breast cancer. While a number of attempts using various image
processing or computer vision techniques have been proposed over the
last decade the current state of the art remains restricted to
specific instances and produces unremarkable results. One particular
challenge is that the collected data is often noisy and contains significant
patient to patient variations.This makes it difficult to train
traditional approaches as collecting labeled data that covers all
possible variation is virtually impossible. Instead, recent advances
in big data related technique promise similar or better results
without the need for labeled training data. More specifically, given
enough data such systems can exploit statistics of large numbers to
reliably detect objects in images. This data set will be of
substantial value to further develop, test, and evaluate such
techniques as a precursor to more wide scale studies.
Aims
1. Adapt big data inspired CT segmentation technique to interpret lung
scans
2. Use the labeled data to evaluate the performance of the algorithms
3. If necessary modify the technique to better fit the use case and
improve performance