Fusing cancer data
to reveal what's
hidden.
MEFINDER integrates radiology, digital pathology, clinical records, and social determinants of health to discover novel cancer phenotypes and deliver population-specific risk predictions — freely, openly, equitably.
260,815+
Breast Cancer Patients
EMBED v2
~1M
Imaging Exams
Multi-modal
5
Institutions
US-wide
10
Open-Source Tools
Freely on GitHub
The cost of one molecular assay — for every patient.
Molecular assays like Decipher are expensive, inconsistently covered by insurance, and still don't capture the full complexity of tumor biology.
Patients with identical diagnoses experience vastly different outcomes. Yet current tools treat them the same. MEFINDER develops computational alternatives that fuse information already collected during routine clinical care.
Current approach
$3,400 / patient
Molecular assay · Limited coverage
MEFINDER
Routine data
Computational · Open-source · Equitable
Same diagnosis.
Different outcomes.
One framework to explain why.
MEFINDER is an NCI U01-funded initiative led by Dr. Judy Gichoya at Emory University's HITI Lab. By integrating five data modalities — radiology, pathology, EHR, genomics, and social determinants of health — the framework discovers novel disease subtypes and predicts who needs more aggressive treatment.
Every tool is open-source, every method reproducible, every dataset documented — so the research community can build on this work rather than starting over.
Read our full missionFive steps from raw data
to clinical insight.
Full Framework 01
Data Collection
Radiology · Pathology · EHR · SDOH
02
Harmonization
DICOM preprocessing · Stain normalization · QC
03
Feature Extraction
Radiomics · Pathomics · Deep embeddings · NLP
04
Multimodal Fusion
Graph networks · Co-attention · Contrastive learning
05
Phenotype Discovery
Novel subtypes · Risk stratification · Treatment guidance
Two cancers. One framework.
MEFINDER targets high-impact prognosis problems where multimodal AI can meaningfully reduce cost and improve equity of care.
ER+ Recurrence Prediction
Integrating mammography, breast MRI, digital pathology, and clinical records from 260,815+ patients to predict late recurrence in ER-positive breast cancer without costly molecular assays.
Biochemical Recurrence
Predicting PSA recurrence post-therapy using biparametric MRI and H&E pathology slides via APIC — delivering Decipher-comparable performance at a fraction of the $3,400 test cost.
10 tools.
All freely available.
From raw DICOM to multimodal embeddings — the full pipeline as open-source software.
HistoQC
stableOpen-source quality control for whole-slide pathology images with thousands of downloads.
View on GitHubF-SYN
stableFourier-based spatial image normalization for stain harmonization — avoids GAN artifacts.
View on GitHubMQUAL
stableMRI quality assessment tool evaluating signal-to-noise, motion artifacts, and sequence completeness.
View on GitHubBeaks
stableCross-modality quality assessment framework for both radiology and pathology image sets.
View on GitHubAPIC
stableAI-based pathology image classifier for tumor-immune interaction and treatment benefit prediction in prostate cancer.
View on GitHubMamoCLIP
stableFederated contrastive learning framework for large-scale mammography representation learning.
View on GitHubFive institutions. One mission.
View all details| Institution | Role | Key Contribution | Datasets |
|---|---|---|---|
| Emory University | Lead | Coordination · HITI Lab · NLP labeling | EMBED v2 · EPIP |
| Indiana University | Pathomics | APIC · Clinical trial validation | CHAARTED · STAMPEDE |
| Stanford | NLP & Breast | NLP toolkit · Data harmonization | Stanford cohort · CA Registry |
| Mayo Clinic | Follow-up | Long-term outcomes (10–15 yr) | Mayo Biobank 75k+ |
| Veterans Affairs | Prostate MRI | Slide digitization · Diverse population | VA bpMRI 387 (expanding) |
Peer-reviewed research.
All publications01
MamoCLIP: A Strong Contrastive Baseline for Full-Field Digital Mammography Analysis
Shrivastava A, Ghosh S, Gichoya JW, et al.
International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI)
02
APIC: AI-Based Pathology Image Classifier for Treatment Benefit Prediction in Prostate Cancer
Bhatt D, Shrivastava A, Bhargava R, Gichoya JW, et al.
Journal of Clinical Oncology: Clinical Cancer Informatics
03
HistoQC: An Open-Source Quality Control Tool for Digital Pathology Slides
Janowczyk A, Zuo R, Gilmore H, Feldman M, Madabhushi A
JCO Clinical Cancer Informatics
DOI: 10.1200/CCI.18.00136The framework is open.
So is the community.
Whether you're a researcher, clinician, patient advocate, or trainee — MEFINDER is built for collaboration.