Detection of Prostate Cancer in 3-Dimensional Pathology Datasets via Generative Immunolabeling

Area of Science:

• Pathology
• Computational Biology
• Oncology

Background:

• Nondestructive 3D pathology enables volumetric analysis of clinical specimens, complementing standard histology.
• Previous 3D pathology studies for prostate cancer risk stratification used manual annotations, lacking fine-grained gland delineation.
• Accurate differentiation between cancerous and benign glands is crucial for precise risk stratification.

Purpose of the Study:

• To automate and enhance the delineation of benign and prostate cancer-enriched regions in 3D pathology datasets.
• To develop a 3D computational pipeline for improved prostate cancer risk stratification.
• To validate the accuracy of the developed pipeline against expert annotations.

Main Methods:

• Developed a 3D computational pipeline named Synthetic Immunolabeling for Generative Heatmaps of Tumor (SIGHT).
• Utilized deep learning-based 3D image translation models for supervised conversion of H&E-analog data to immunofluorescence datasets.
• SIGHT synthetically labels cytokeratin markers for generating 3D heatmaps of cancer-enriched regions.

Main Results:

• SIGHT achieved an average F1 score of 0.88 in validation against expert annotations, comparable to inter-pathologist agreement (F1 score of 0.90).
• Preliminary machine classifiers using SIGHT-identified features showed a significant hazard ratio of 3.57 for recurrence risk.
• Volumetric glandular analysis in SIGHT-identified regions significantly improved prostate cancer risk stratification compared to analysis of all tissue regions.

Conclusions:

• SIGHT effectively automates the delineation of cancerous and benign prostate glands in 3D pathology, enhancing accuracy.
• The pipeline facilitates improved prostate cancer risk stratification through precise analysis of 3D histomorphometric features.
• SIGHT represents a significant advancement in leveraging 3D pathology for personalized cancer diagnostics and prognostics.