Landing Page
Initial Page for simulation
Brain 3d simulation

Synovia

Inspiration

Neurosurgeons remove 500,000 brain tumors per year and 20% of surgeries have unexpected complications because surgeons can't predict tissue behavior. Traditional pre-operative mapping takes hours and only shows current function - not what happens when tissue is removed. We asked: "What if AI could predict surgical consequences in real-time?"

What it does

Synovia predicts neurological consequences of brain tissue removal in under a 10 seconds using AI.

Input: Brain region, coordinates, volume to remove, patient age
Output: Predicted deficits (motor, language, cognitive), recovery timeline, surgical approach recommendations, risk probabilities

Neurosurgeons can test multiple "what if" scenarios instantly before making the first incision - like a flight simulator for brain surgery.

ML & Simulation Backend (FastAPI + Python):

Physics-driven FEA Engine: Implemented finite element analysis (FEA) to simulate stress propagation, strain tensors, and displacement fields in cortical regions following virtual resections.

Material Modeling: Incorporated viscoelastic tissue parameters and nonlinear elasticity constants from neurophysiology literature to ensure biomechanical realism.

Gemini 2.0 Integration: Leveraged Google Gemini’s multimodal reasoning for real-time neurological interpretation—mapping FEA-derived stress fields to potential post-surgical deficits and recovery profiles.

Multi-stage ML Pipeline: Combined voxel-wise CNN-based segmentation (for 2D→3D reconstruction) with physics-informed neural networks (PINNs) for mechanical field prediction.

Normalized Coordinate Framework: Designed a spatial reference system aligning MRI voxel space with Talairach coordinates for precise lesion and FEA localization.

Data Validation & Schema Design: Built Pydantic-based integrity checks for volumetric and biomechanical inputs, ensuring stable end-to-end model serving.

API Orchestration: FastAPI routes for /upload, /segment, /simulate, and /fea endpoints handle DICOM ingestion, NIfTI processing, and mesh export pipelines.

🧩 3D Visualization Frontend (Next.js + Three.js):

Interactive Brain Reconstruction: Converted segmented 3D meshes (.STL) into dynamically rendered Three.js geometries, enabling surgical region selection and deformation visualization in real-time.

Physics-Aware Rendering: Integrated vertex color gradients mapped to stress tensor magnitudes and principal strain vectors for intuitive biomechanical insight.

Coordinate Mapping Engine: Real-time projection of simulation coordinates onto the anatomical mesh—synchronizing physics outputs with interactive model state.

WebSocket-Driven Feedback Loop: Enabled continuous updates from FEA solver to front-end for real-time visualization of stress redistribution during virtual resection.

Shader-Enhanced Display: Custom GLSL shaders for material translucency, lighting realism, and cortical depth cues to better illustrate internal deformation patterns.

End-to-End Integration: Seamless pipeline from 2D MRI upload → 3D reconstruction → physics-based FEA → Gemini-powered interpretation—all accessible through a single web interface.

Challenges we ran into

Some of the challenges we ran to included finding a problem space we all cared for and a solution we wanted to build out, and merging some of our features to make sure they were compatible and connected.

Accomplishments that we're proud of

Physics-driven modeling: Integrated real-world biomechanical physics into our simulations to accurately predict post-surgical brain stress and deformation. Our FEA-based engine mimics true tissue responses, not just visual approximations.

2D MRI → 3D Brain Reconstruction: Built a full physics and ML-powered segmentation pipeline that converts 2D MRI slices into anatomically precise 3D models—enabling dynamic visualization and region-specific stress mapping.

ML-powered intelligence: Leveraged Gemini’s large-scale reasoning and fine-tuned ML inference to interpret 3D outputs, identify high-risk neural zones, and generate explainable insights for neurosurgical planning.

ML: Trained reasoning using Gemini API to do analysis.

Seamless integration: Built a complete end-to-end pipeline from 3D brain visualization to AI-powered predictions.

What we learned

We learned about the importance about persistence and gained more knowledge about how to convert 2D scans to 3D models.

What's next for Synovia

Patient-specific MRI upload and automated brain segmentation
Historical outcome database - train on real neurosurgical cases
Multi-region scenarios - predict combined resection impacts
Voice-guided planning with ElevenLabs for hands-free operation during surgery