1:30-3:00pm

1. Standardizing protocol/protocol alignment for multi-information biomarker algorithms - Grogan

a) Align and standardized (as best and responsible possible) biospecimen collection and storage, patient level data (core sets of demographics, risk factors, clinical history, etc.), and medical image (e.g., image acquisition parameters, image processing, imaged de-dentification, etc).

b) If successful above, EDRN has the resources and expertise to develop multi-information biomarker algorithms (e.g., imaging + circulating + risk factors + ?) for all cancers of interest in the current and future EDRN portfolio.

c) Development and multisite validation of high throughout risk/diagnosis calculators for cancers that integrate clinical, imaging, biomarker data for different cancers

Discussion Points:

1. What are existing clinical data infrastructure collection mechanisms (REDCap, DMCC data structure etc.. that could be used for different projects
2. Standard demographics and disease specific risk factors
3. Clinical utility measures (like unnecessary biopsies in prostate), time to diagnosis of cancer, clinical decisions such as f/u, imaging or biopsy that should be stored
4. Existing cloud based imaging storage that allow images to reside at institutions and be called up for specific radiomic projects
5. Standard EDRN tissue/biospecimen protocols for each disease that allow each site to store samples in same manner.

2. Federated learning (FL) project (images ± patient data ± ‘omic data) - Schabath

a) Federated learning (FL) is a paradigm shifting solution to mitigate the bottleneck of training and validating large volumes of PHI data (and does not violate privacy)

b) Individual sites can train on their data allow without violating privacy issues.

c) By adding enhanced privacy preservation and uncertainty estimation, FL provides a viable solution for large scale machine learning deployments in the medical imaging domain (but not exclusive to just medical imaging)

Discussion Points:

1. There are potential new infrastructure costs for give sites (GPUs, curation of data, research databases to house the curated data)
2. How much data standardization is needed?
3. There are numerous difference FL methods, privacy preservation methods, and uncertainty estimations: This requires experimentation to reveal advantages, barriers, limitations, etc.
4. Each site will likely need CS/Data science expertise to manage the infrastructure
5. Using FL for medical imaging is considered the standard use case, there has been little consideration for other types of data (large scale ‘omic data, EMR data, pathology images, etc)
6. How do garner interest in a multi-site FL project?

3. Use of existing biomarkers in the EDRN for clinical utility trials - Marquez

a) Use of Test cancer-specific assays in other cancer sites

Assays that were trained and validated on a specific cancer (e.g., head and neck), may be relevant and perform well in a different cancer (e.g., lung).

Discussion Points:

1. What are the bars the EDRN assays need to meet in order to move to a CUT?
2. Would EDRN or CSRN (Cancer Screening Research Network) guide the CUT?
3. If CSRN, what role would the EDRN PI have, if any? Protocol development, data analysis

b) Commercial biomarkers and Pan-biomarkers/MCDs/MCEDs

Partner with commercial companies (and non-commercial entities) that have MCD/MCED platforms or organ-specific assays and test in EDRN studies.

Discussion Points:

1. What role would EDRN play? feasibility, preliminary data, pilot?

4. Biomarkers for recurrence – POTENTIAL OTHER TOPIC

EDRN has largely been diagnosed on risk, early detection, and diagnosis. Yet, the EDRN has expanded to include cancer recurrence (and death) in early-stage disease. So, testing existing biomarker algorithms (or develop new biomarker algorithms) for cancer recurrence in early-stage and/or screen detected cancers.

a) Mathematical models for natural trajectory from risk à diagnosis à clinical outcomes

Mathematical Oncology has emerged as an important area of research that integrate applied and theoretical mathematics with experimental data to better understand cancer initiation, progression, and treatment and to aid in the clinical utilization of integrated models in precision medicine. These principles have yet to be fully appreciated in the early detection setting.

b) Likewise, develop biomarker algorithms for indolent disease?