CT provides structure
Computed tomography contributes anatomy: bone, airway, soft tissue boundaries and spatial context.
Multimodal Oncologic Imaging
OncoFusion AI
3D PET/CT Head & Neck Tumor Segmentation
A presentation-ready research workflow that combines anatomical CT structure with metabolic PET activity to delineate gross tumor volume across a standardized 3D scan.
Powered with ❤️ by Solicitous ®
3Dvolumetric analysis
PET + CTmultimodal input
144³standardized ROI
Fusion Review / Case HN-042
AI READY
Active modalityPET/CT Fusion
Voxel grid1 mm isotropic
Review stateSegmentation visible
Metabolic hotspot
GTV contour
The clinical logic
What is actually happening?
The model is not simply highlighting a bright spot. It aligns two complementary imaging signals, normalizes them, focuses on the head-and-neck region, and predicts a 3D gross-tumor-volume mask.
PET provides activity
Positron emission tomography shows radiotracer uptake, helping reveal metabolically active tumor tissue.
Fusion provides context
The network learns where uptake sits relative to anatomy, reducing reliance on either modality alone.
The output is a volume
Instead of one image-level label, the output is a voxel-level tumor mask spanning the scan volume.
Interactive modality theatre
Normalization changes what the network sees
Switch between CT, raw PET and two PET normalization strategies. The animation preserves the original panels and overlays a scan cursor without cropping the medical image.
GTV
Anatomical reference for localization
End-to-end workflow
From scanner data to a reviewable tumor volume
STEP 01 / DATA INGEST
Load paired PET and CT studies
The workflow starts with co-registered imaging volumes. CT provides structural anatomy while PET contributes metabolic activity.
Smart section
PET normalization laboratory
Normalization can suppress distracting uptake and reshape intensity contrast. This matters because a segmentation model learns from the visual distribution it receives.
Clipping strategyLimits extreme uptake values so the tumor contour receives stronger relative contrast.
Face-mask filteringUses a head-and-neck mask to reduce irrelevant activity outside the anatomical region of interest.
Transformation simulatorPET Clip
Clipping compresses extreme hotspots while preserving a strong lesion-to-background contrast.
Visual evidence
Same patient, different preprocessing, different contour behavior
The project figures compare expert ground truth with predictions from baseline PET and transformed PET channels. Values shown are example-case Dice scores from the supplied figures, not a universal clinical performance claim.
Smart interpretation
For this compact lesion, clipping improves contour agreement in the supplied example by compressing the competing high-intensity hotspot.
Model engine
3D segmentation core
The source workflow uses nnU-Net as the baseline framework for full-resolution 3D gross-tumor-volume segmentation and evaluates models through five-fold cross-validation.
CT
PET
+feature fusion
ENCODER
3D latent
representation
representation
DECODER
GTV Maskvoxel-wise output
Why 3D?Tumors extend across slices; volumetric context helps maintain continuity.
Why multimodal?PET and CT answer different questions and become more useful when interpreted together.
Why cross-validation?Five-fold evaluation tests stability across multiple train/validation splits.
Beyond segmentation
Imaging features can support treatment-outcome research
The source project also contains a separate treatment-outcome prediction track. Segmentation can define the tumor region from which radiomic or learned features are extracted for downstream research such as progression-free-survival prediction.
Research support — not a diagnostic or prognostic device
1Segment tumor
→2Extract features
→3Build model
→4Review risk signal
Instrument-connected vision
How this becomes a compelling local demo
Scanner ingestion
Load paired PET and CT studies from a local demonstration folder.
Automatic preprocessing
Resample, crop and normalize with visible progress and quality checks.
Interactive contour review
Inspect the mask over CT, PET and fusion views before accepting it.
Quantitative report
Summarize tumor volume, slice span, centroid and modality observations.
OncoFusion AI Analysis SummaryDemonstration report · Case HN-042
RESEARCH USEDetected volume12.8 cm³Illustrative UI value
Slice coverage31 slicesAxial extent
Primary modalityPET/CTFused interpretation
Review statusPendingExpert confirmation required
AI-assisted multimodal oncology imagingSee anatomy. See activity.
Return to Top ↑See anatomy. See activity.
Understand the volume.
Powered with ❤️ by Solicitous ®