Study
PLCO
(Learn more about this study)
Project ID
PLCOI-1539
Initial CDAS Request Approval
Apr 29, 2024
Title
Development and Validation of In-House Artificial Intelligence Models for Mutation and Survival Prediction
Summary
The aim of the proposed analysis is threefold. Firstly, we plan to independently validate pre-existing AI models optimized with H&E stained whole tumor sections to perform a) binary mutation classification for several genes (i.e. TP53, PIK3CA, BRCA1-2), b) calculate post-processed image biomarkers (TILs), and c) validate pre-existing AI models optimized for survival prediction. To this end, PLCO’s breast subset will be used as an external cohort to validate the predictive performance and investigate the prognostic value of the AI models and their corresponding calculated covariates/biomarkers. Secondly, we will also exploit phase 1 covariates and developed AI models from our group to calculate and integrate pathology-based descriptors towards increasing prognostication and prediction in PLCO setting. Thirdly, we plan on utilizing PLCO breast subset together with TCGA and other in-house cohorts to further train the AI models for increased performance and generalizability. Such AI models have the potential to enhance clinical practice by offering more accurate prognostic insights and personalized risk assessments based on pathology-derived data. This tailored approach could assist clinicians in refining treatment strategies, potentially optimizing patient outcomes. Additionally, their improved generalizability might make them more accessible across various healthcare settings, potentially aiding clinicians in making informed decisions to improve patient care universally.
Aims
1. Prognostic validation of in-house models: Validate models trained for mutation classification (TP53, PIK3CA, BRCA1-2), image biomarker calculation (TILs), and survival prediction using PLCO breast subset in terms of their prognostic performance.
2. Integrate Pathology-Based Descriptors: Use phase 1 covariates and developed AI models to integrate pathology-based descriptors for improved prognostication in PLCO setting.
3. Enhance Models and Generalizability: Further train AI models using PLCO, TCGA, and other in-house cohorts to increase performance and generalizability for clinical use.
Collaborators
Kang Wang, Karolinska Institutet, Sweden (kang.wang@ki.se)
Associate Professor Theodoros Foukakis, Karolinska Institutet, Sweden (theodoros.foukakis@ki.se)
