Multi-Modal Deep Learning for Predicting Screening Outcome from Imaging and Structured Data
Principal Investigator
Name
Rebecca Simpson
Degrees
B.Eng
Institution
J&B Engineering
Position Title
Research Engineer
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-472
Initial CDAS Request Approval
Jan 24, 2019
Title
Multi-Modal Deep Learning for Predicting Screening Outcome from Imaging and Structured Data
Summary
The interest and developments in deep learning applied to medical imaging have resulted in many improvements for tasks such as organ segmentation and lesion detection. Although these approaches tend to focus purely on imaging, there is often additional data, including medical survey results, blood tests, clinical notes in both structured or free form text which could also provide insights and guidance to these deep learning models. Often this data is not included due to architecture constraints and the perception that structured data is best handled using traditional machine learning models. This project will investigate a comparison between including and excluding this data as part of a deep learning architecture in terms of how accurately it can predict the follow-up screening results of a patient given their first screening lung CT. There will also be a comparison of machine learning approaches to predicting screening outcome for the first scan against more recent novel deep learning approaches to predictions on structured data.
Aims
* Assess whether a deep learning approach can yield greater accuracy than traditional machine learning ones when predicting initial screening outcome prior to imaging
* Assess whether a deep learning method designed to predict the outcome of follow-up screening events performs better when lung screening survey data is included compared to when it is not included.
Collaborators
N/A