1
Upload
Provide a CT scan volume and let the platform prepare it for analysis.
Clinical-grade AI review
Detect lung cancer earlier with AI-powered analysis.
Calm, focused software for CT scan upload, nodule detection, and risk review.
OncoVision-X transforms complex chest CT volumes into clear findings, actionable next steps, and a professional results workflow built for serious medical review.
How it works
One focused workflow from upload to review.
2
Analyze
The AI pipeline detects nodules, calculates malignancy probability, and assembles timing metrics.
3
Review
Inspect the scan visualization, risk banner, recommendations, and past results in a structured workspace.
Trust
Built to feel clear, serious, and medically grounded.
Features
A platform, not a single tool.
Fast analysis
Focused upload-to-results workflow with clean state transitions.
Detailed reports
Risk banner, nodule cards, CT views, and export-ready scan summaries.
Secure organization
Review past scans from a dedicated history page instead of losing context in one long screen.
Export options
Generate image exports today, with report presets scaffolded into the interface.
Ready to analyze your first scan?
Get StartedOR
Analyzing CT Scan...
Preparing volumetric data
12% complete
Scan Results
Detailed findings, timing, and recommendations.
File
-
Size
-
Format
-
CT visualization
Three-view scan review
Visualization will appear here when the backend returns CT overlays.
High Risk
Medium Risk
Low Risk
Your Scan History
0 scans analyzed
Behind the Technology
Deep dive into the model architecture, validation work, performance, and developer-facing details.
Architecture
DCA-Net: Dual-Context Attention Network
The production pipeline combines local nodule evidence with wider anatomical context, then fuses both streams through attention before classification.
Input CT Scan
↓
Preprocessing & Candidate Detection
↓
Nodule Branch
64×64×64
64×64×64
Context Branch
48×48×48
48×48×48
Dual Attention Fusion
↓
Detection + Classification
Key components
- 3D ResNet feature extraction for nodule-centric evidence.
- Wider context branch to reduce false positives from vessels and scar tissue.
- Attention fusion that learns local/global weighting per candidate.
Training details
- Dataset: LUNA16-based split with full-model evaluation artifacts in this repository.
- Framework: PyTorch pipeline with attention fusion and uncertainty outputs.
- Outputs: nodule locations, malignancy probabilities, and scan-level risk summaries.
Validation
Research & ablation results
This section uses the repository’s experiment outputs directly rather than placeholder graphics.
Full DCA-Net0.958 AUC0.891 F1
No Attention0.932 AUC0.854 F1
No Context0.915 AUC0.837 F1
ResNet3D-180.887 AUC0.803 F1
Performance
Detection metrics and operating characteristics
Performance views below use experiment artifacts already generated in the repository.
For developers
Developer resources and integration direction
The current app exposes analysis through the Flask endpoint in this repository. The public API described here is future-facing, but grounded in the existing request shape.
Current app endpoint
curl -X POST http://localhost:5000/api/analyze \
-F "ct_scan=@patient_scan.nii.gz"Python example
import requests
response = requests.post(
"http://localhost:5000/api/analyze",
files={"ct_scan": open("patient_scan.nii.gz", "rb")},
)
print(response.json()["analysis"])Open source direction
- Training and evaluation artifacts already live in this repository.
- Experiment plots can be embedded into reports, validation pages, and documentation.
- The frontend now exposes a dedicated technical surface for researchers and collaborators.
Future API surface
POST /api/v1/analyzefor job submission.GET /api/v1/results/:jobIdfor completed outputs.- Key-based auth, audit logging, and report generation as follow-on work.