SURF: Announcements of Opportunity
Below are Announcements of Opportunity posted by Caltech faculty and JPL technical staff for the SURF program. Additional AOs for the Amgen Scholars program can be found here.
Specific GROWTH projects being offerred for summer 2018 can be found here.
Each AO indicates whether or not it is open to non-Caltech students. If an AO is NOT open to non-Caltech students, please DO NOT contact the mentor.
Announcements of Opportunity are posted as they are received. Please check back regularly for new AO submissions! Remember: This is just one way that you can go about identifying a suitable project and/or mentor.
Announcements for external summer programs are listed here.
Students pursuing opportunities at JPL must be
U.S. citizens or U.S. permanent residents.
|Project:||Classifying Lung Nodules Using Machine Learning|
|Disciplines:||Data Science, Computer Science|
Sr. Research Scientist, (PMA),
|Mentor URL:||http://www.astro.caltech.edu/~aam (opens in new window)|
|Background:||Early Detection Research Network (EDRN) and Molecular and Cellular Characterization of Screen-Detected Lesions (MCL) are two large NIH funded cancer research programs and networks where JPL data science center plays a key role. Using best practices developed for the complex, diverse and voluminous planetary science datasets, many datasets related to these cancer studies are organized in a manner that makes novel investigations feasible. The deidentified datasets include various images, RNA sequences, time-series and so on. Given the advances in machine learning, even more studies can be arranged around these datasets that would not have been possible until recently. One such dataset is that of 3D low-dose computed tomography (LDCT) lung images of smokers obtained as part of the National Lung Screening Trial (NLST).|
|Description:||There are two components to the project. One involves creating a labeled dataset from a set of images that contain lung nodules. A tutorial is being constructed with the help of MCL Imaging Working Group lead by Denise Aberle at the University of California, Los Angeles to label the consistency, shapes, margins, and other features of nodules. A larger set of images with nodules will then be presented to trained radiologists for creating the training set using a citizen science like approach based on zooniverse. We will then use the even larger NLST dataset to identify and classify lung nodules. The identification will involve first segmenting the images, and the classification part will involving running deep learning codes using standard convolutional neural network libraries. In the past feature-based classification has been carried out on such image sets. Typically a large fraction of nodules found are false positives (and there is a very small fraction of false negatives as well, typically interval cancers). We will directly compare the performance of deep learning with the traditional methods, and also use the false positives known from traditional methods to improve classification. If time permits we will use multiple scans as a time-series to understand predictability.|
Deep Learning: http://deeplearning.net/tutorial/
Pinsky et al., Ann Intern Med. 2015;162(7):485-491
Gillies RJ, Kinahan PE, Hricak H., 2016, Radiology. 278, 2, 563
Schabath et al., 2016, PLOS ONE, 11(8): e0159880
|Student Requirements:||Proficiency in python, conversant with statistics basics, deep learning, using GPUs and AWS, knowledge about linux/unix, git and related software engineering techniques. Basic bioinformatics knowledge will be a plus.|
This AO can be done under the following programs:
<< Prev Record 35 of 151 Next >> Back To List