Unsupervised Machine Learning to discover new COPD subtypes

We are sorry, this position has been filled.

In 5R01HL121270 our team successfully applied graph-based unsupervised machine learning to spatial and visual features of lung parenchyma in HRCT scans, independent of all clinical information, which resulted in the discovery of six new CT quantitative emphysema subtypes, five of which bear marked similarities to the classic but discarded COPD subtypes and three of which had specific genetic associations with biologically plausible genes (paper under review at The Lancet). The current project will apply similar graph-based unsupervised machine learning to airway trees, to define quantitative trees (Qutees) subtypes in the general population.

Lab: Heffner Biomedical Imaging Laboratory

Direct Supervisor: Andrew Laine

Position Dates: 6/1/2020 - 8/31/2020

Hours per Week: 20

Paid Position: Yes

Credit: No

Number of Positions: 2

Qualifications: Programming in Python and C desired.

Eligibility: Sophomore, Junior, Senior, Master's (SEAS only)


Andrew Laine, laine@columbia.edu