Skip to main content
Research

Research.

Published work, validation methodology, and the science behind DART's diagnostic models.

Published Research
Conference Paper

An Agentic AI Framework for Unified Brain Health Profiling: Integrating Multimodal Neurological Data with Foundation Models

Conference of Applied Scientific & Machine Learning 2025

Indian Institute of Science (IISc), Bengaluru

This paper presents DART's approach to multi-modal neurological data fusion — combining EEG, VNG, MRI, and clinical records into a unified diagnostic output running entirely on-premise. Validated across 97+ neurologists in the US, India, Switzerland, and Canada.

View / Download Whitepaper
Venue
IISc Bengaluru · 2025
Status
Published
Team
Braindata Intelligence Pvt. Ltd.

Deeper Problem Understanding

The inefficiency is structural. Not a workflow problem — a data architecture problem.

01

Fragmented Inputs

EEG, VNG, ANS, CNSVS, Posturography, imaging & PDFs live in separate systems. A neurologist spends 6 billable hours generating a single med-legal report.

205 neurologists & attorneys validated: avg 4–6 hrs / report

02

Missed Cross-Modal Signals

Human bandwidth cannot reliably link signals across tests and patient history. High-impact diagnostic cues are routinely missed.

56% of TBI deaths in the US result from misdiagnosis

03

No Comprehensive Patient Profile

Neurologists lack time to build a holistic patient picture. Signal-to-noise ratio is critically high in complex neurological cases.

Confirmed by attorneys, insurers & clinicians alike

Validation Summary

Before Review #1

97+

Neurologists across US, India, Switzerland, Canada

Cognitive load than normal in med-legal TBI cases

Since Review #1 — New

85

New conversations post-R1

3

Audience types engaged

Neurologists · Attorneys · Device companies

Key post-R1 insight

Device companies see DART as an embedded analytics layer.

Positive Impact in MVP

Measured outcomes from live deployments across 3 hospitals and 8 law offices.

Saving Critical Time for Neurologists

Diagnostic TaskBeforeTime NowReduced
Cognitive & Limbic Vitals45 mins3 mins93%
EEG Calculations & Interpretations3 hours4 mins95%
Videonystamography Interpretations60 mins2 mins93%
Vestibular & Gait Interpretations20 mins2 mins90%
1,000-page Medical Records Review2 days4 hours99%
Radiology Imaging Analysis3+ hours3 mins> 99%

Uncovering Previously Hidden Insights

We were able to link depression/anxiety to longer saccade latencies and show frontal-theta's role in saccade control in 85% of outputs of patients with depression history.

In an MVA dispute, MedRecs flagged healthy pre-accident neurology. An independent neurologist confirmed. Outcome: $25,000 case won for a single mother.

When minutes mattered

“At 2 a.m., MedSight flagged TBI patterns on a Brain MRI. I urged the night OPD for immediate treatment — the decision helped save my friend's mother.”

How DART Works

DART's models run entirely on-premise inside the client's infrastructure. No patient data is transferred or stored externally under any circumstances.

The system fuses EEG, VNG, ANS, CNSVS, Posturography, imaging, and clinical PDF records into a single synthesised output using a multi-modal fusion pipeline. Each data stream is normalised before fusion to eliminate format inconsistency as a source of diagnostic error.

Every claim in the output is anchored to a source document or diagnostic finding. The report includes reason codes and citations so that clinicians and attorneys can trace any conclusion back to its evidence.

The long-term goal is a longitudinal Brain Credit Score — a standardised, explainable neurological profile for every patient, built incrementally across appointments and updated as new data arrives.

On-prem / VPCAll inference runs on the client's own hardware. Nothing leaves the hospital or law office network.
HIPAA-alignedArchitecture-level compliance. RBAC, audit trails, and access controls built into the Dartboxes layer.
RAG + CAGRetrieval-augmented and context-augmented generation for document intelligence in MedRecs.
DICOM-compatibleMedSight reads standard medical imaging formats directly — no conversion step required.
Multi-modal fusionEEG, VNG, ANS, CNSVS, Posturography, MRI, and clinical PDFs fused in a single pipeline.
TRL Level 5System validated in an operational environment — 3 hospitals and 8 law offices, Washington State.
Mistake score: 6/100Down from 8/100 at launch. Measured across 1,672 report instances.
Avg output: 7 minAcross 1,672 report instances (as of Mar 15, 2026). Previous baseline: 6 billable hours per neurologist.

Want to see the full methodology behind a DART report?
Request a pilot call and we will walk through a real output on your data.

Book a Pilot Call