THE FUTURE OF ALZHEIMER'S
EARLY DETECTION
AI-powered 3-tier screening that catches risk before symptoms appear.
Why MirAI?
Traditional Alzheimer's diagnosis relies on expensive MRI, PET, or invasive CSF tests—only accessible
after cognitive decline begins.
MirAI changes this. We use accessible clinical, genetic, and blood-based biomarkers to
detect risk before symptoms appear.
Stage 1: Clinical
Non-invasive screening using memory questionnaires and functional assessments.
Stage 2: Genetic
Risk stratification using APOE ε4 and polygenic markers.
Stage 3: Biomarker
Blood-based pTau-217, Aβ42, and NfL analysis.
The Problem We're Solving
55 million people worldwide live with dementia. By 2050, this will triple. Yet diagnosis typically occurs only after significant cognitive decline—when treatment options are limited.
-
Delayed Detection
Traditional diagnostics (MRI, PET, CSF) are expensive, invasive, and only used after symptoms appear.
-
Scalable Solution
MirAI uses accessible inputs—questionnaires, genetic tests, and simple blood draws—enabling population-level screening.
-
Clinically Sound
Our 3-tier escalation prevents over-testing while ensuring high-risk individuals receive proper follow-up.
Million Patients Worldwide
Stage Escalation Pipeline
% Stage 3 AUC Accuracy
ADNI Patients Analyzed
How It Works
MirAI's 3-Tier Escalation Pathway ensures accurate risk assessment with minimal unnecessary testing.
Initial Assessment
Non-invasive, low-cost indicators including:
- Subjective memory complaints
- Functional activities (FAQ score)
- Age and family history
- Everyday cognition (ECog)
This stage deliberately avoids diagnostic scales (MMSE, CDR) to prevent label leakage and ensure valid early detection.
Risk Refinement
Individuals flagged in Stage 1 undergo genetic risk scoring:
- APOE ε4 - Strongest genetic risk factor
- Polygenic risk markers
APOE ε4 association with late-onset AD was established by Allen D. Roses.
Biochemical Confirmation
High-risk individuals are assessed through plasma biomarkers:
- pTau-217 - Most accurate plasma marker
- pTau217/Aβ42 ratio
- Neurofilament Light (NfL)
Clinical validity of plasma pTau217 established by Oskar Hansson's research.
Ready to Assess Your Risk?
Our AI-powered screening takes just 5 minutes and provides instant, transparent results with explainability.
Start ScreeningThe Science
Explore the technical details behind each stage of MirAI's screening pipeline.
Clinical Screening Model
XGBoost classifier trained on ADNI data
Features: AGE, PTGENDER, PTEDUCAT, FAQ, EcogPtMem, EcogPtTotal
Pipeline: KNN Imputer → Standard Scaler → XGBoost
Performance: AUC 0.87 on held-out test set
Uses cross-validation predictions (not direct training predictions) to prevent stacking leakage when passing probabilities to Stage 2.
Genetic Risk Stratification
Risk refinement using genetic susceptibility
Features: Stage1_Prob, APOE4_Count
Key Insight: APOE ε4 acts as a "Risk Modifier" rather than a sole determinant—it refines the clinical signal.
Performance: AUC 0.88 (marginal improvement over Stage 1 alone)
Biomarker Confirmation
Blood-based pathology detection
Features: Stage2_Prob, pT217_F, AB42_F, AB40_F, NfL_Q
Key Finding: pTau-217 is the dominant biomarker—considered the most specific plasma marker for AD pathology.
Performance: AUC 0.93 (significant improvement over previous stages)
Integrated Decision Engine
Cascade fusion with explainability
Architecture: Each stage passes calibrated probabilities to the next, ensuring uncertainty propagation.
Leakage Prevention: Patient-level separation using GroupShuffleSplit on RID ensures no patient appears in both train and test.
Output: Final 0-100% risk score with stage-by-stage breakdown and feature importance.
Research Foundation
MirAI's methodology is grounded in decades of pioneering Alzheimer's research.
Marshal Folstein
MMSE DeveloperDeveloped the Mini-Mental State Examination (MMSE), the most widely used cognitive assessment. MirAI does NOT use MMSE as input—we treat it as a diagnostic tool (not a screening tool) to prevent data leakage and ensure true early detection.
Allen D. Roses
APOE ε4 PioneerFirst demonstrated the association between APOE ε4 and late-onset Alzheimer's Disease, establishing it as the strongest genetic risk factor for sporadic AD.
Oskar Hansson
pTau-217 ResearcherLed pivotal studies establishing plasma pTau-217 as one of the most accurate blood-based biomarkers for detecting Alzheimer's pathology.
Frequently Asked Questions
Common questions about MirAI's early detection approach.
Is MirAI a diagnostic tool?
No. MirAI is a screening tool, not a diagnostic. It identifies individuals at elevated risk who should then undergo formal clinical evaluation. A positive result does NOT mean you have Alzheimer's—it means further assessment is recommended.
Why don't you use MMSE or CDR scores?
MMSE and CDR are diagnostic instruments designed to assess existing cognitive impairment. Using them in a screening model creates "label leakage"—the model would essentially be predicting the diagnosis from diagnosis-adjacent features, defeating the purpose of early detection.
How accurate is the biomarker stage?
Stage 3 achieves an AUC of approximately 0.93 on held-out test data. However, accuracy alone doesn't tell the full story—MirAI also provides uncertainty estimates and stage-by-stage explainability so clinicians can make informed decisions.
What data does MirAI use?
MirAI was trained on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, which includes clinical assessments, genetic markers (APOE), and plasma biomarker data from over 2,400 participants.
Is my data stored or shared?
In this demo version, all processing happens locally in your browser or on a local server. No personal health data is transmitted to external servers. For production deployment, all data handling would comply with HIPAA/GDPR requirements.
Contact
Get in touch with the Break&Build team.