PrePCa: Prostate Cancer Prediction Based on Multimodal Data
Principal Investigator
Name
Maria Dech Pons
Degrees
M.Sc.
Institution
Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.
Position Title
Researcher Data Science & Biostatistics
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-1760
Initial CDAS Request Approval
Dec 9, 2024
Title
PrePCa: Prostate Cancer Prediction Based on Multimodal Data
Summary
Our project is planned in two steps:
1) predict the prostate cancer ISUP grade score using a combination of different - to this moment - tabular data, including broad clinical information, genetic data, questionnaires and results from clinical examinations.
2) given the gleason score and knowledge about histopathological findings, recommend the best therapy.
Key Research Questions:
- To what extent can existing logistic regression models predict the ISUP grade / Gleason score of prostate cancer (PCa)?
- Which features are the most salient for prediction?
- Are base models biased towards populations with certain properties?
- Can alternative approaches, such as machine learning or ensemble models, achieve higher predictive accuracy for ISUP grade scores?
- Which features are needed to give a good therapy recommendation?
Project Approach:
We are currently compiling existing models to evaluate their validity using external datasets, such as the PLCO dataset for prostate. Most of these models have not been externally validated. Our goal is to develop a new model that integrates the strengths of existing models and identifies the conditions under which each performs optimally.
In addition, we aim to collect supplementary data to enhance cross-center robustness and gain deeper insights into the learning patterns of current models.
To facilitate the application of these models, we require access to comprehensive datasets with covariables covering personal, genetic and clinical data collected up until biopsy; and from biopsy on for the second phase of the project.
1) predict the prostate cancer ISUP grade score using a combination of different - to this moment - tabular data, including broad clinical information, genetic data, questionnaires and results from clinical examinations.
2) given the gleason score and knowledge about histopathological findings, recommend the best therapy.
Key Research Questions:
- To what extent can existing logistic regression models predict the ISUP grade / Gleason score of prostate cancer (PCa)?
- Which features are the most salient for prediction?
- Are base models biased towards populations with certain properties?
- Can alternative approaches, such as machine learning or ensemble models, achieve higher predictive accuracy for ISUP grade scores?
- Which features are needed to give a good therapy recommendation?
Project Approach:
We are currently compiling existing models to evaluate their validity using external datasets, such as the PLCO dataset for prostate. Most of these models have not been externally validated. Our goal is to develop a new model that integrates the strengths of existing models and identifies the conditions under which each performs optimally.
In addition, we aim to collect supplementary data to enhance cross-center robustness and gain deeper insights into the learning patterns of current models.
To facilitate the application of these models, we require access to comprehensive datasets with covariables covering personal, genetic and clinical data collected up until biopsy; and from biopsy on for the second phase of the project.
Aims
- Validate existing predictive models on external datasets.
- Identify the most information-rich features for predicting prostate cancer ISUP grades to deduce a more efficient screening pipeline and offer a robust risk stratification for biopsy.
- Develop a robust, integrative model leveraging all available data and models
- Identify salient features for a reliable therapy recommendation
Collaborators
- Max Westphal - Fraunhofer Institut, MEVIS, Bremen, Germany
- Horst Hahn - Fraunhofer Institut, MEVIS, Bremen, Germany