Scientific Blog
Pathology AI Insights
Technical articles on IHC scoring methodology, validation statistics, digital pathology integration, and computational oncology from the Synthia team.
USCAP 2026 Annual Meeting: Key Themes in Digital Pathology and IHC AI
A summary of the digital pathology and computational IHC themes from the USCAP 2026 Annual Meeting — including emerging validation standards and SaMD regulatory discussions.
AI in Pathology: The Case for Augmentation, Not Automation
The framing of "AI replacing pathologists" is counterproductive. Pathologist judgment is irreplaceable. What AI can — and should — do is eliminate the tedium of manual quantification.
Combining Convolutional and Attention Mechanisms for Precise IHC Cell Detection
Hybrid convolutional-attention models achieve better cell localization in crowded IHC-stained fields — a look at the technical approach and why pure CNN architectures struggle with WSI.
HER2 2+ Equivocal Cases: How Automated Flagging Reduces Unnecessary ISH Reflex Testing
Equivocal HER2 2+ cases create uncertainty and trigger expensive ISH reflex testing. Algorithmic confidence scoring helps prioritize which 2+ cases are borderline vs. clearly progressing toward 3+.
Where Pathology Labs Lose Hours: A Workflow Audit of IHC Turnaround Time
We mapped the IHC reporting workflow at three academic cancer centers. The biggest bottleneck is not staining or scanning — it's the manual scoring step taking 30–90 minutes per slide set.
Building High-Quality Training Data for Pathology AI: The Annotation Consensus Problem
The quality of a pathology AI model is only as good as its training annotations. Multi-reader adjudication protocols, discordance resolution, and why "average of three pathologists" is harder than it sounds.
Navigating the 510(k) Pathway for IHC Quantification Software as a Medical Device
FDA's SaMD framework has evolved significantly. For IHC biomarker scoring tools, here is what the regulatory landscape looks like in 2025 and what documentation standards apply.
Ki-67 Hotspot Detection: Why Spatial Context Changes the Proliferation Index
Ki-67 scoring isn't just counting positive cells across a slide — it's identifying the highest-density proliferative region. Our approach to hotspot localization and why global vs. hotspot scores tell different stories.
Integrating Whole-Slide Imaging with LIS and EHR Systems: A Technical Overview
Academic medical centers face real infrastructure friction when adopting digital pathology tools. We map the integration landscape — DICOM-WSI, REST APIs, HL7 FHIR, and what lab IT teams need to know.
PD-L1 Scoring Across Assays: 22C3, 28-8, and SP142 Are Not Interchangeable
PD-L1 immunohistochemistry uses at least three commercial antibody clones approved for companion diagnostics — each with different scoring thresholds. Algorithmic PD-L1 scoring must account for this.
ICC and Weighted Kappa: What These Statistics Actually Tell You About AI Pathology Performance
ICC and weighted kappa are the gold-standard metrics for evaluating diagnostic AI in pathology. Here is what they measure, how they differ, and what thresholds actually matter clinically.
Why Inter-Pathologist HER2 Scoring Variability Remains a Clinical Challenge
Manual HER2 IHC scoring in breast cancer biopsy review has a documented inter-observer variability rate that affects treatment eligibility decisions. We examine the data.