Skip to Main Content

COVID-19 is an emerging, rapidly evolving situation.

What people with cancer should know: https://www.cancer.gov/coronavirus

Get the latest public health information from CDC: https://www.coronavirus.gov

Get the latest research information from NIH: https://covid19.nih.gov/

Principal Investigator
Name
Gil Shamai
Degrees
Ph.D.
Institution
Technion
Position Title
Postdoc
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-727
Initial CDAS Request Approval
Nov 17, 2020
Title
Predicting survival and molecular expression from histological images in lung cancer
Summary
Hematoxylin and Eosin (H&E) is the basic pathological specimen staining and is routinely done for biopsy samples. H&E allows pathologists to examine the morphology of the tissue under a microscope in order to evaluate whether it is malignant and characterize the type and grade of the tumor. Nevertheless, the information that is extracted from such a procedure, when traditionally analyzed manually by pathologists, is limited. Pathologists cannot evaluate from H&E staining, for example, the molecular profile of the cancer, expression of biomarkers, and gene mutations. Such characteristics require advanced techniques, such as immunohistochemistry (IHC), fluorescent in situ hybridization (FISH), and DNA sequencing. These techniques may be expensive and time consuming, and their interpretation contains inter-observer variability. Moreover, today, these technologies do not exist in every country.

We believe that computational learning systems have the potential of revealing new and yet unexploited information in tumor morphology. In our preliminary study, we showed that a deep learning-based system was able to accurately predict the ER status of breast cancer patients from H&E-stained histological images.
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2739045.
We plan to scale up this research in order to assess the potential of artificial intelligence to predict survival time and hormonal receptors expression (such as EGFR, ALK, and PDL-1) from H&E-stained images in lung cancer patients.

We envision the proposed technology to play a pivotal role in workflow analysis and targeted therapy. Similar to our preliminary study for receptor status prediction, survival time in lung cancer could potentially be accurately predicted. For those patients who would be classified with high morphological signals by the system, molecular identification using direct assays may be improved and even unnecessary.
Aims

1) Predict receptor status for different biomarkers in lung cancer patients, based on image analysis and convolutional neural networks.
2) Predict survival times in lung cancer patients, based on image analysis and convolutional neural networks.

Collaborators

Gil Shamai - Technion
Ron Kimmel - Technion