StrokeDense-CTA: Offline Clinical Emergency Neuro-Radiology Foundation Engine for Multi-Phase CT Angiography Large Vessel Occlusion & Collateral Scoring AI
Advancing open-source emergency neuro-radiology infrastructure, StrokeDense-CTA delivers verified clinical decision support, deterministic safety gating, and high-throughput offline edge execution.
PedsCardio-EchoFlow: Offline Clinical Pediatric Cardiology Foundation Engine for Congenital Heart Disease Echocardiography Doppler AI
Advancing open-source pediatric cardiology infrastructure, PedsCardio-EchoFlow delivers verified clinical decision support, deterministic safety gating, and high-throughput offline edge execution.
CytoMap-Leukemia: Offline Clinical Clinical Hematology Foundation Engine for High-Dimensional Flow Cytometry Automated Gating & MRD Quantification AI
Advancing open-source clinical hematology infrastructure, CytoMap-Leukemia delivers verified clinical decision support, deterministic safety gating, and high-throughput offline edge execution.
RenalGraft-VXM: Offline Clinical Transplant Nephrology Foundation Engine for Renal Allograft Rejection & Single-Cell Transcriptomic Biomarker Engine
Advancing open-source transplant nephrology infrastructure, RenalGraft-VXM delivers verified clinical decision support, deterministic safety gating, and high-throughput offline edge execution.
UniMedVL: Unified Multimodal Foundation Model for Cross-Modal Clinical Understanding & Diagnostic Reasoning
Structured around an Observation-Knowledge-Analysis paradigm, UniMedVL enables synchronized visual question answering, automated structured radiology report drafting, and cross-modal diagnostic reasoning.
Prov-GigaPath: Whole-Slide Pathology Foundation Model Pre-Trained on 1.3 Billion Parameters from Real-World Biopsies
Trained on 171,189 whole-slide biopsy images encompassing 31 cancer subtypes, Prov-GigaPath establishes a new state of the art in mutation prediction, cancer subtyping, and survival modeling with unprecedented zero-shot transferability.
MedGemma: Google's Open-Weight Foundation Models for Clinical Reasoning, Medical Imaging, and Genomics
A deep architectural and clinical evaluation of the MedGemma model family, covering 2B, 9B, and 27B parameter checkpoints with native medical imaging encoders and EHR-sequence fine-tuning capabilities.
TxGNN: Geometric Deep Learning on Knowledge Graphs Identifies Novel Therapies for 17,080 Diseases
Harvard's TxGNN leverages medical knowledge graphs to benchmark zero-shot drug repurposing across thousands of conditions without dedicated drug development pipelines, including rare pediatric and metabolic disorders.
GenomeTransformer: Deep Generative Modeling Predicts Cryptic Splicing Mutations and Rare Variant Clinical Effects
A comprehensive analysis of multi-scale genomic foundation models trained on cross-species sequence conservation to predict non-coding variant consequences and cryptic splice site creation in rare disease cohorts.
EchoCLIP: Contrastive Multi-Modal Foundation Model for Automated Echocardiogram Interpretation and Hemodynamic Assessment
Pre-trained on over 1.2 million cardiac ultrasound cine loops and linked sonographer reports, EchoCLIP demonstrates expert-level chamber quantification, ejection fraction estimation, and structural valve disease classification.
PathoFoundation: Self-Supervised Vision Transformer for Pan-Cancer Gigapixel Whole-Slide Image Analysis and Biomarker Discovery
Evaluating hierarchical token aggregation and masked image modeling across 28 organ sites to achieve unprecedented diagnostic accuracy for rare histopathologic subtypes and biomarker expression forecasting.
The Open-Source Frontier: Inside the 619 Models, Cookbooks & Tools Powering the OpenPHR Catalog
A comprehensive audit of the open-source medical AI ecosystem, spanning clinical NLP, medical imaging foundation models, genomic variant prediction, and local privacy-preserving edge architectures.