How mandeep15686/mask-r-cnn-lung-nodule-detection-edge-cloud-pipeline works
This research prototype combines LUNA16 Mask R-CNN training and evaluation with a FastAPI service, React dashboard, and edge–cloud experiments. A typical workflow is to prepare cached 2.5D patches, run a configured training experiment, inspect its metrics and leaderboard, then upload a patch for checkpoint-backed inference and review the saved result. The README also describes federated learning, split inference, compression, and privacy research. Several routes and edge internals were not sampled, so their connections are omitted or marked unverified.
Main entry points: main.py FastAPI app · App.jsx App navigation · trainer.py Trainer · api.js HTTP and WS client
Components 32
Research dashboard
- App navigation
frontend/src/App.jsx - Dashboard
frontend/src/pages/Dashboard.jsx - Experiments UI
frontend/src/pages/Experiments.jsx - Inference UI
frontend/src/pages/InferencePage.jsx - Edge network UI
frontend/src/pages/EdgeNetwork.jsx - Leaderboard
frontend/src/pages/Leaderboard.jsx - HTTP and WS client
frontend/src/services/api.js
API and persistence
- FastAPI app
backend/app/main.py - Experiment routes
backend/app/routes/experiments.py - Inference routes
backend/app/routes/inference.py - Results routes
backend/app/routes/results.py - Edge routes
backend/app/routes/edge.py - Dashboard routes
backend/app/routes/dashboard.py - Inference engine
backend/app/ml/inference.py - Experiment and result DB
backend/app/database.py - Data configuration
backend/app/data_config.py
Training and evaluation
- Patch cache
training/preprocess_cache.py - LUNA16 dataset
training/dataset.py - Trainer
training/trainer.py - Mask R-CNN model
training/model.py - Mask losses
training/losses.py - Evaluation metrics
training/metrics.py - Phase orchestration
training/experiment.py
Edge-cloud research
- Federated simulation
training/federated.py - Edge node service
edge/edge_node.py - Split feature extractor
edge/feature_extractor.py - Feature compressor
edge/compressor.py - Network monitor
edge/network_monitor.py - Edge preprocessing
edge/preprocessor.py
Research utilities
- Split analysis
scripts/cn1_split_analysis.py - Privacy experiments
scripts/cn5_differential_privacy.py
Other components
- Researcher
Connections 44
- Researcher → App navigation: uses
frontend/src/App.jsx - App navigation → Dashboard: routes to
frontend/src/App.jsx - App navigation → Experiments UI: routes to
frontend/src/App.jsx - App navigation → Inference UI: routes to
frontend/src/App.jsx - App navigation → Edge network UI: routes to
frontend/src/App.jsx - App navigation → Leaderboard: routes to
frontend/src/App.jsx - Dashboard → HTTP and WS client: requests data
frontend/src/pages/Dashboard.jsx - Experiments UI → HTTP and WS client: requests experiments
- Inference UI → HTTP and WS client: submits patches
frontend/src/pages/InferencePage.jsx - Edge network UI → HTTP and WS client: requests network data
- Leaderboard → HTTP and WS client: requests rankings
- HTTP and WS client → FastAPI app: sends HTTP and WS
- FastAPI app → Inference routes: registers router
backend/app/main.py - FastAPI app → Experiment routes: registers router
backend/app/main.py - FastAPI app → Results routes: registers router
backend/app/main.py - FastAPI app → Edge routes: registers router
backend/app/main.py - FastAPI app → Dashboard routes: registers router
backend/app/main.py - FastAPI app → Experiment and result DB: creates tables
backend/app/main.py - FastAPI app → Inference engine: initializes
backend/app/main.py - FastAPI app → Data configuration: reads runtime info
backend/app/main.py - Inference routes → Inference engine: runs inference
backend/app/routes/inference.py - Inference routes → Experiment and result DB: stores and reads results
backend/app/routes/inference.py - Experiment routes → Experiment and result DB: manages experiments
backend/app/routes/experiments.py - Experiment routes → Trainer: starts training
backend/app/routes/experiments.py - Experiment routes → Data configuration: resolves data paths
backend/app/routes/experiments.py - Trainer → Mask R-CNN model: trains
training/trainer.py - Trainer → LUNA16 dataset: loads batches
training/trainer.py - Trainer → Mask losses: computes losses
training/trainer.py - Trainer → Evaluation metrics: evaluates
training/trainer.py - LUNA16 dataset → Patch cache: reads cached patches
training/dataset.py - Patch cache → LUNA16 dataset: uses dataset helpers
training/preprocess_cache.py - Inference engine → Mask R-CNN model: loads checkpoint model
backend/app/ml/inference.py - Phase orchestration → Trainer: orchestrates phases
- Federated simulation → Mask R-CNN model: uses model
training/federated.py - Federated simulation → LUNA16 dataset: uses data
training/federated.py - Federated simulation → Mask losses: uses losses
training/federated.py - Federated simulation → Evaluation metrics: uses metrics
training/federated.py - Edge node service → FastAPI app: communicates with cloud
- Edge node service → Split feature extractor: uses split inference
- Edge node service → Feature compressor: uses compression
- Edge node service → Network monitor: measures network
- Edge node service → Edge preprocessing: prepares inputs
- Split analysis → Split feature extractor: studies split points
- Privacy experiments → Federated simulation: explores privacy