Identifying histological features of lung cancer using deep learning

Principal Investigator

Name
Jennifer Beane

Degrees
Ph.D.

Institution
Trustees of Boston University

Position Title
Associate Professor

Email
jbeane@bu.edu

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-1494

Initial CDAS Request Approval
Jan 26, 2026

Title
Identifying histological features of lung cancer using deep learning

Summary
Our group is interested in developing deep learning frameworks to learn histologic patterns and features in lung tissue and their associated microenvironment. We have developed different deep learning methods that can be applied to H&E whole slide images to predict lung cancer and lung cancer subtypes that we will leverage to build predictors of more detailed histologic features. We would like to gain access to the NLST pathology images to develop predictors of prognosis-associated histologic features that we can test in our institution cohorts. Through collaboration with thoracic pathologists at our institution, the slides will be scored for specific histologic features of interest, and we will train predictive models using the NLST data. We plan to test the performance of these models in highly curated institutional cohorts and associate the model-derived features with genomic features available on the same samples.

Aims

1. Score NLST H&E digitized slides for pathologic features of interest
2. Develop robust deep learning models to predict the scored pathologic features using the NLST H&E digitized slides and evaluate model performance in independent cases from our institution where multiple slides per case are available. Test each model's reliance on the pathologic feature by assessing performance in cases where the feature is present in the tumor but not in the slide, where the features is present in the tumor and in the slide, and where the feature is absent.
3. Identify gene expression associated with model-derived features across the independent cases using bulk RNA-seq and spatial transcriptomic data.

Collaborators

Jennifer Beane Boston University School of Medicine
Vijaya Kolachalama Boston University School of Medicine