PLCO Digital Pathology Whole-Slide Images for Prostate, Lung, and Ovarian Cancer AI Modeling and Validation

Principal Investigator

Name
Yuming Jiang

Degrees
M.D., Ph.D.

Institution
Wake Forest University Health Sciences

Position Title
Tenure Track Assistant Professor

Email
yuming.jiang@wfusm.edu

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCOI-2052

Initial CDAS Request Approval
Jun 8, 2026

Title
PLCO Digital Pathology Whole-Slide Images for Prostate, Lung, and Ovarian Cancer AI Modeling and Validation

Summary
We request access to PLCO prostate, lung, and ovarian digital pathology whole-slide images (WSI) to support the WSI branch of our multimodal AI research program. In this CDAS Images project, we will request and use pathology images. Linkage to participant-level phenotype, clinical, and outcome tables will be handled separately through a companion CDAS Data-Only project, as appropriate and after separate approval.

WSIs will be processed with our established computational pathology pipeline, including tissue masking and tiling, stain normalization, artifact quality control, nuclei segmentation and classification, and extraction of cell-level and tissue-level image features. We will also extract foundation-model WSI embeddings and use attention-based multiple instance learning to generate slide-level and patient-level image representations. These image representations will support downstream modeling and validation after appropriate approval for matched clinical or outcome data.

This project focuses on image resource acquisition, image preprocessing, feature extraction, representation learning, and cross-cohort robustness analysis. We will evaluate image quality, tissue morphology, nuclear phenotypes, and cancer-site-specific image patterns across prostate, lung, and ovarian cancer WSIs. When external public cohorts are available, we will compare image representations and preprocessing robustness against external datasets such as TCGA-based cohorts.

No re-identification will be attempted, and WSIs will not be redistributed. Data will be stored on institution-approved secure servers with encryption at rest and access limited to approved users. All analyses will report aggregate results only and will follow CDAS data use and publication policies.

Aims

Aim 1: Harmonized WSI acquisition and preprocessing.
We will request and organize PLCO prostate, lung, and ovarian digital pathology WSIs. We will standardize tissue masking, image tiling, stain normalization, artifact filtering, and quality control under versioned preprocessing configurations. Deliverables will include organized image manifests, preprocessing logs, and reproducible configuration files.

Aim 2: Cell-level phenotyping and interpretable image feature extraction.
We will implement nuclei segmentation and classification to extract cell-level morphology, spatial organization, and tissue microenvironment features from WSIs. Additional hand-crafted pathology features and CellProfiler-derived nuclear features will be extracted when feasible. Deliverables will include tile-level and slide-level feature matrices, data dictionaries, and quality assurance summaries.

Aim 3: Foundation-model WSI representation learning.
We will extract deep pathology image embeddings from tiles and aggregate them into slide-level and patient-level WSI representations using attention-based multiple instance learning. This aim will generate reusable image representations for prostate, lung, and ovarian cancer WSIs that can support future approved multimodal modeling. Deliverables will include embedding files, model cards, and reproducible feature extraction notebooks.

Aim 4: Cross-cancer image pattern analysis and external robustness testing.
We will compare WSI-derived representations across prostate, lung, and ovarian cancer to evaluate shared and cancer-site-specific morphology patterns. Under the same preprocessing pipeline, we will assess robustness using external public pathology cohorts when available, including TCGA-based datasets. Deliverables will include cross-cancer representation analyses, reliability plots, and external robustness reports.

Aim 5: Secure data management and CDAS compliance.
All PLCO images and derived files will be stored on institution-approved secure servers with access limited to approved project personnel. We will not attempt participant re-identification, contact participants, or redistribute images. We will maintain access logs, follow the approved data use period, and complete project close-out and certified data destruction according to CDAS policy.

Collaborators

Yuming Jiang Wake Forest University Health Sciences
GUANNAN HE Wake Forest University Health Sciences
Yijun Chen Wake Forest University Health Sciences