ProstAId-CDS - A Clinical Decision Support System for Prostate Cancer Treatment
Principal Investigator
Name
Stefan Bonn
Degrees
Ph.D.
Institution
University Medical Center Hamburg Eppendorf / ZMNH
Position Title
Professor, Institute Director
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-589
Initial CDAS Request Approval
Mar 27, 2020
Title
ProstAId-CDS - A Clinical Decision Support System for Prostate Cancer Treatment
Summary
Summary: The central aim of the ProstAId project is to develop a clinical decision support system (CDS) for Prostate Cancer (PCa) therapy by utilizing cutting edge AI algorithms and knowledge bases. Previous large-scale studies in PCa treatment (such as ERSPC or the PLCO) have refined statistical predictive methods for prostate cancer therapy. Some of these statistical models, so called nomograms, are in active use as clinical decision support systems (for an overview, see [1],[2]).
While nomograms prove the usability of in silico approaches to support clinical decision making in PCa, they rely on the integration of few parameters and do not make use of the power of recent developments in Deep Learning. We aim to use state-of-the-art machine- and deep-learning (ML and DL) techniques on large and diverse PCa clinical data to build a next generation of high-performance clinical decision support systems (CDS). To this end, we collaborate with the local prostate clinic (the bAIOme and the Martiniklinik at the University Medical Center Hamburg Eppendorf, Germany), who possess well-documented patient data starting in 1992 (with close to 30,000 radical prostatectomies performed). We will integrate information from patient records including follow-up data (average of 60 months), pathology and MRI fusion data, and build a custom knowledge database designed to incorporate data from other clinics or clinical studies.
By using patient records and images from different clinics and/or clinical studies, we want to optimize the performance of our algorithm and address the heterogeneity of data of differents cohorts or clinics. We will use a federated model learning approach to address this bias together with a number of European partners, thus ensuring the CDS stability and generalisability necessary to bring a Smart Medical Expert System into every-day clinical use (at the University Medical Center Hamburg-Eppendorf and beyond).
For validation and benchmarking of our trained model(s) we are seeking additional external data sets of high quality and a high degree of standardization. For this purpose, access to PLCO data would be of tremendous value to our project.
[1] Vinod P. Balachandran, Mithat Gonen, J. Joshua Smith, MD, and Ronald P. DeMatteo. Nomograms in Oncology – More than Meets the Eye. Lancet Oncol. (2015), e173-e180, doi: 10.1016/S1470-2045(14)71116-7
[2] Alex Z. Fu, Scott B. Cantor, Michael W. Kattan. Use of Nomograms for Personalized Decision-Analytic Recommendations. Med Decis Making. (2010), 30(2):267-74. https://doi.org/10.1177/0272989X09342278
While nomograms prove the usability of in silico approaches to support clinical decision making in PCa, they rely on the integration of few parameters and do not make use of the power of recent developments in Deep Learning. We aim to use state-of-the-art machine- and deep-learning (ML and DL) techniques on large and diverse PCa clinical data to build a next generation of high-performance clinical decision support systems (CDS). To this end, we collaborate with the local prostate clinic (the bAIOme and the Martiniklinik at the University Medical Center Hamburg Eppendorf, Germany), who possess well-documented patient data starting in 1992 (with close to 30,000 radical prostatectomies performed). We will integrate information from patient records including follow-up data (average of 60 months), pathology and MRI fusion data, and build a custom knowledge database designed to incorporate data from other clinics or clinical studies.
By using patient records and images from different clinics and/or clinical studies, we want to optimize the performance of our algorithm and address the heterogeneity of data of differents cohorts or clinics. We will use a federated model learning approach to address this bias together with a number of European partners, thus ensuring the CDS stability and generalisability necessary to bring a Smart Medical Expert System into every-day clinical use (at the University Medical Center Hamburg-Eppendorf and beyond).
For validation and benchmarking of our trained model(s) we are seeking additional external data sets of high quality and a high degree of standardization. For this purpose, access to PLCO data would be of tremendous value to our project.
[1] Vinod P. Balachandran, Mithat Gonen, J. Joshua Smith, MD, and Ronald P. DeMatteo. Nomograms in Oncology – More than Meets the Eye. Lancet Oncol. (2015), e173-e180, doi: 10.1016/S1470-2045(14)71116-7
[2] Alex Z. Fu, Scott B. Cantor, Michael W. Kattan. Use of Nomograms for Personalized Decision-Analytic Recommendations. Med Decis Making. (2010), 30(2):267-74. https://doi.org/10.1177/0272989X09342278
Aims
• Proof-of-concept: Design of an AI-based smart clinical decision support system for crucial decisions in prostate cancer therapy at the local clinic
• Increase the accuracy and robustness of the ML/DL-models by extension to a large amount of data of project partners via federated learning
• validation and benchmarking on external data (PLCO)
• devise a Smart Medical Expert System for every-day clinical use
Collaborators
IMSB (ZMNH, University Medical Center Hamburg-Eppendorf): Stefan Bonn
Martiniklinik (University Medical Center Hamburg-Eppendorf): Lars Budäus
Siemens/Ludwig-Maximilians-Universität Munich: Volker Tresp