Towards personalized cancer treatment with deep learning
Principal Investigator
Name
Francesco Ciompi
Institution
Radboud University Medical Center
Position Title
Senior researcher
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-323
Initial CDAS Request Approval
Nov 24, 2017
Title
Towards personalized cancer treatment with deep learning
Summary
In this project, we aim at designing computer models based on deep learning [Good17, LeCu15, Schm15] to support physicians in the decision of an optimal, personalized treatment for cancer patients.
Deep learning models will process digitized histopathology images and will be trained to predict clinically relevant information, solely based on the HE-stained digitized specimens.
In particular, targets for the computer model will be:
* classification of tumor type and molecular subtypes
* assessment of patient prognosis
* prediction of adjuvant treatment response
* assessment of several imaging biomarkers (e.g., tumor grading based on mitotic activity, tumor-stroma ratio, amount of tumor-infiltrating lymphocytes)
Histopathology images will be used to train computer models based on deep learning, and the patient and clinical data will define the target that will be used to train the models.
Data of lung, colorectal and breast cancer will be used in this project.
Deep learning models will process digitized histopathology images and will be trained to predict clinically relevant information, solely based on the HE-stained digitized specimens.
In particular, targets for the computer model will be:
* classification of tumor type and molecular subtypes
* assessment of patient prognosis
* prediction of adjuvant treatment response
* assessment of several imaging biomarkers (e.g., tumor grading based on mitotic activity, tumor-stroma ratio, amount of tumor-infiltrating lymphocytes)
Histopathology images will be used to train computer models based on deep learning, and the patient and clinical data will define the target that will be used to train the models.
Data of lung, colorectal and breast cancer will be used in this project.
Aims
The automatic quantification of imaging biomarkers to aid defining optimal personalized treatments for cancer patients.
Collaborators
Jeroen van der Laak, Computational Pathology Group, Radboud University Medical Center Nijmegen, Netherlands
Oscar Gessink, Computational Pathology Group, Radboud University Medical Center Nijmegen, Netherlands