PAI-WSIT

PAI-WSIT aggregates and annotates whole-slide images (WSIs) and applies deep learning analyses to detect malignant regions and classify phenotypic subtypes for tumor characterization and precision oncology research.

Key Features:

• Data Collection and Integration: A standardized dataset comprising 8,633 WSIs from 1,772 tumor cases with a primary focus on colorectal cancer (CRC) sourced from four regional hospitals in China and The Cancer Genome Atlas (TCGA), plus breast, lung, prostate, bladder, and kidney cancers from two Chinese hospitals.
• Annotation Capabilities: An integrated multifunctional annotation tool enabling high-precision slide annotations and collaborative review of pathologist annotations.
• AI-Powered Analysis: Use of deep learning frameworks to detect malignant regions and to classify phenotypic subtypes in colorectal cancers, including identification of subvisual morphometric phenotypes.
• Performance Metrics: Reported WSI classification accuracy of 0.933 for malignant region detection in CRC and an area under the curve (AUC) of 0.719 on colorectal subtype datasets.
• Comprehensive Data Repository: Inclusion of gene detection reports from 582 tumor cases and extensive clinical information for all recorded cases.

Scientific Applications:

• AI-assisted tumor diagnosis: Integration of molecular detection data with WSI analysis to support identification of malignant regions and diagnostic decision support.
• Precision oncology and biomarker development: Mining of morphometric phenotypes to inform predictive assays and translational research for patient stratification.
• Colorectal cancer research: Automated detection and subtype classification in CRC WSIs, with abnormal-region visualization via heatmaps to aid interpretability.

Methodology:

Application of deep learning frameworks for WSI classification, malignant-region detection, and colorectal subtype classification, with abnormal-region visualization represented as heatmaps.