Skip to content

ROADMAP.md

Latest commit

 

History

History
260 lines (204 loc) · 7.67 KB

ROADMAP.md

File metadata and controls

260 lines (204 loc) · 7.67 KB

SRODecoderRing - Project Roadmap

This document outlines planned improvements and development priorities for the SRODecoderRing project.

Current Status

The project currently has:

  • ✅ Working Energy-Based Model (EBM) for rotation order inference
  • ✅ PyTorch-based training pipeline
  • ✅ F# production inference engine
  • ✅ Flask REST API deployed on Azure
  • ✅ Elmish web frontend on IPFS
  • ✅ WPF desktop application

Known Issues (Immediate Attention)

High Priority

  1. Site Association (README.md:182)

    • Multi-site treatment planning needs refinement
    • Current implementation may not handle complex multi-site cases correctly
    • Impact: Patient treatment accuracy

  2. iView Field Association (README.md:183)

    • Issue: Applying shifts after first field with TPO that won't associate with second field
    • Impact: Multi-field treatment workflows
    • Needs investigation and resolution

Short-Term Improvements (Q1-Q2 2025)

Documentation & Developer Experience

  • API Documentation

    • Add OpenAPI/Swagger spec for Flask REST API
    • Document request/response formats
    • Provide example API calls with curl/Python/JavaScript

  • Training Guide

    • Step-by-step tutorial for training new models
    • Hyperparameter tuning guide
    • Dataset preparation best practices

  • Deployment Guide

    • Document IPFS deployment process
    • Azure deployment automation
    • Docker containerization for Flask API

  • Code Documentation

    • Add docstrings to Python modules (numpy/Google style)
    • XML documentation comments for C#/F# code
    • Inline comments for complex algorithms

Testing & Quality Assurance

  • Expand Test Coverage

    • Python unit tests for EBM components
    • Integration tests for API endpoints
    • UI end-to-end tests with Selenium/Playwright

  • Continuous Integration

    • GitHub Actions workflow for automated testing
    • Automated build verification for .NET components
    • Python linting (ruff, mypy, black)

  • Clinical Validation Suite

    • Known test cases from clinical data
    • Regression test suite
    • Performance benchmarks

Model Improvements

  • Model Performance

    • Benchmark inference time (target: <100ms)
    • Optimize MCMC sampling efficiency
    • Investigate model quantization for faster inference

  • Training Infrastructure

    • Add training metrics dashboard (TensorBoard/Weights & Biases)
    • Automated hyperparameter search
    • Model versioning and experiment tracking

Medium-Term Enhancements (Q3-Q4 2025)

Feature Development

  • Enhanced Multi-Field Support

    • Resolve iView field association issue
    • Support complex multi-field treatment plans
    • Handle field-specific rotation orders

  • Site Association Improvements

    • Implement robust multi-site treatment planning
    • Add site-specific coordinate system handling
    • Validate against clinical scenarios

  • Uncertainty Quantification

    • Provide confidence scores for predictions
    • Alert when model is uncertain
    • Human-in-the-loop review for low-confidence cases

  • Real DICOM Integration

    • Direct DICOM SRO parsing
    • Integration with DICOM network (DIMSE)
    • Support for DICOM worklist queries

Architecture Enhancements

  • Microservices Architecture

    • Separate training service
    • Dedicated inference service
    • Model registry service

  • Caching Layer

    • Redis cache for frequent queries
    • Model weight caching
    • Response caching with invalidation

  • Observability

    • Structured logging (JSON)
    • Distributed tracing (OpenTelemetry)
    • Metrics and alerting (Prometheus/Grafana)

User Experience

  • Web UI Improvements

    • Batch processing interface
    • Historical query visualization
    • Export results to CSV/Excel

  • Desktop UI Enhancements

    • Real-time visualization of transformations
    • Interactive 3D coordinate system viewer
    • Drag-and-drop DICOM file support

  • Mobile Support

    • Responsive web design
    • Progressive Web App (PWA)
    • Offline capability

Long-Term Vision (2026+)

Advanced ML Capabilities

  • Active Learning

    • Collect user feedback on predictions
    • Retrain model with corrected examples
    • Continuous model improvement

  • Multi-Task Learning

    • Predict rotation order + patient position
    • Joint learning of coordinate systems
    • Transfer learning from related tasks

  • Explainable AI

    • Attention visualization
    • SHAP values for interpretability
    • Human-readable explanations

Clinical Integration

  • Treatment Planning System Integration

    • Plugin for Eclipse/RayStation/Pinnacle
    • Native integration APIs
    • Real-time validation during planning

  • Clinical Decision Support

    • Alert system for unusual transformations
    • Automated quality assurance checks
    • Integration with clinical workflows

  • Regulatory Compliance

    • FDA 510(k) preparation (if pursuing medical device status)
    • IEC 62304 software lifecycle compliance
    • Clinical validation studies

Research & Innovation

  • Lie Group Mathematics

    • Leverage SO(3) structure explicitly
    • Geometric deep learning approaches
    • Equivariant neural networks

  • Federated Learning

    • Train on distributed clinical datasets
    • Privacy-preserving learning
    • Multi-institutional collaboration

  • Benchmarking

    • Public dataset for rotation order prediction
    • Comparison with analytical methods
    • Published research paper

Infrastructure

  • Multi-Cloud Support

    • Support AWS, GCP in addition to Azure
    • Kubernetes deployment
    • Auto-scaling inference

  • Edge Deployment

    • ONNX export for edge devices
    • TensorRT optimization
    • Embedded systems support

Migration from Keras to PyTorch

Status: ✅ Complete

The project has migrated from Keras to PyTorch 2.0.1+. Legacy references in code/docs should be updated:

  • Core training pipeline (completed)
  • Remove Keras references from comments
  • Update any remaining Keras-based notebooks

Performance Targets

Metric Current Target (2025) Target (2026)
Inference Time ~200ms <100ms <50ms
Model Accuracy ~95% >98% >99.5%
API Availability 99% 99.9% 99.95%
Test Coverage ~60% >80% >90%

Contributing to the Roadmap

Have ideas for improvements? Here's how to contribute:

  1. Open an Issue: Describe the feature or improvement
  2. Discuss: Engage with maintainers and community
  3. Prioritize: Help us understand impact and urgency
  4. Implement: Submit a PR once approved

Timeline Summary

2025 Q1-Q2: Documentation, Testing, CI/CD
    │
    ├─ API docs & deployment guides
    ├─ Expanded test coverage
    └─ Fix known issues (site association, iView)

2025 Q3-Q4: Features & Architecture
    │
    ├─ Multi-field support
    ├─ DICOM integration
    ├─ Microservices architecture
    └─ Enhanced UI/UX

2026+: Advanced ML & Clinical Integration
    │
    ├─ Active learning & explainability
    ├─ Treatment planning system plugins
    ├─ Regulatory compliance
    └─ Research publications

Version Milestones

  • v1.0: Fix critical issues, comprehensive docs, 80% test coverage
  • v1.5: Multi-field support, DICOM integration, <100ms inference
  • v2.0: Microservices architecture, uncertainty quantification, TPS integration
  • v3.0: FDA clearance path, federated learning, clinical validation

Feedback

This roadmap is a living document. Priorities may shift based on:

  • Clinical feedback and requirements
  • Technical discoveries and blockers
  • Community contributions
  • Regulatory requirements

Last updated: 2025-11-09