The Fifth Elephant 2025 Annual Conference CfP

Problem

Neurodegenerative disorders like Alzheimer’s and Parkinson’s are on the rise, yet early detection remains elusive. Conventional imaging reports often lack quantitative insights and longitudinal baselines, limiting clinicians’ ability to identify subtle neuroanatomical changes over time. A reliable biomarker—brain age—can offer an interpretable and clinically relevant measure of brain health, but building such tools requires robust segmentation, normative data modeling, and population-specific calibration.

Challenges in Existing Brain MRI Analytics

• Lack of Standardized Quantitative Metrics: Current neuroimaging workflows rarely provide structure-wise volumetric percentile analysis, making interpretation subjective and inconsistent.
• Generalized Models Trained on Non-Local Populations: Many brain age models are trained on non-representative populations, leading to limited applicability to Indian datasets.
• Insufficient Data Infrastructure: Most clinical systems lack curated datasets and tooling to compare new scans against healthy, demographically matched cohorts.

Solution

We present a population-calibrated, volumetric analytics and brain age estimation framework for brain MRI analysis:

• Built on nn-UNet, a state-of-the-art deep learning architecture trained to segment 132 brain structures from T1-weighted MRIs with ~5,000 high-quality 3D brain MRIs.
• A curated normative database of ~1,300 high-quality 3D brain MRIs from Indian patients forms the foundation for population-specific benchmarking.
• Each new scan is segmented, volumetrically analyzed, and compared against this normative reference to calculate percentiles for each brain structure, contextualized by age and gender.
• Using these volumes, a regression-based brain age prediction model calibrated on Indian data estimates deviations from expected aging patterns, enabling early detection of neurodegenerative risk.

Agenda

1. Introduction
   Framing the burden of neurodegenerative disorders and the need for early detection using quantitative imaging.

2. Gaps in Current Brain MRI Analysis
   Why existing reports lack actionable metrics, and the risks of non-local modeling in brain age prediction.

3. Segmentation and Normative Database Design
   Training nn-UNet on local data for structure-level brain segmentation, and building an age- and gender-indexed volume percentile database.

4. Volumetric Percentile Analytics
   How each new scan is benchmarked against normative data, generating structure-wise percentile maps.

5. Brain Age Modeling
   Using volumetric data to predict brain age, interpret brain age gaps, and explore associations with neurodegenerative risk.

6. Results and Insights
   Demo: From DICOM to risk-aware brain age report – a walkthrough of the end-to-end pipeline.

7. Future Directions and Collaboration Opportunities
   From model refinement to clinical validation: where the project can grow, and how collaborators can contribute.

Key Takeaways for Attendees

Attendees will leave with a deep understanding of:

• How population-specific volumetric benchmarking can enhance brain MRI interpretation.
• The potential of brain age as a biomarker for early neurodegeneration.
• How to build practical, scalable segmentation and analytics systems grounded in local data.

Speaker

Sandhiya CV – Data Scientist, 5C Network
Sandhiya is a Data Scientist specializing in Computer Vision, Deep Learning, and Vision-Language Models (VLMs) for medical imaging, with a focus on Radiology. She designs and deploys scalable, end-to-end AI pipelines that integrate multi-modal data to improve diagnostic accuracy, efficiency, and clinical decision-making. Her work delivers interpretable, robust, and clinically relevant solutions at the intersection of AI and healthcare.

Slide Deck
Predicting Brain Age GAP