Validation of a Self-Supervised Model for Tumor Recognition and Generalizability using the PLCO Cancer Screening Trial Dataset
Principal Investigator
Name
Addie Dvir
Degrees
M.Sc
Institution
Imagene
Position Title
Head of Product
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-1158
Initial CDAS Request Approval
Feb 8, 2023
Title
Validation of a Self-Supervised Model for Tumor Recognition and Generalizability using the PLCO Cancer Screening Trial Dataset
Summary
An increase in the adoption of digital scanning technologies paves the way for the modernization of the pathology lab and the development of diagnostic tools based on artificial intelligence and deep learning (AI/DL). We have developed models that decipher tumor regions from benign, alert for intrinsic artifacts, and predict occult biomarkers in H&E whole slide images (WSI). With the help of PLCO datasets, we wish to increase generalizability and validate a tumor-specific segmentation model that can assist the diagnosis process.
Aims
1. Train our segmentation model on multiple tumor types including lung, breast, prostate, colon, ovarian, and others.
2. Develop inter-tumor histology segmentation by self-supervised learning. For example, adenocarcinoma vs. squamous in lung cancer, ductal vs. lobular in breast cancer, and grade differentiation.
3. Use of convolutional neural network (CNN) for patterns of the prognosis within tumor groups.
Collaborators
No external collaborators