Automating X-Ray Workflows: Raspberry Pi, GitHub Actions, and MorphoSource Integration

1. The Need for Scalable and Secure X-Ray Metadata Automation

The world of X-ray imaging encompasses an array of devices, each tailored for specific applications. A snapshot of commonly used scanners highlights the diversity of brands and models:

These numbers represent just a sample of the extensive usage scenarios and underscore the heterogeneous landscape of X-ray machinery. When examining typical metadata from repositories like MorphoSource, we can see the complexity of fields that need to be captured:

These numbers represent just a sample of the extensive usage scenarios and underscore the heterogeneous landscape of X-ray machinery, but it is apparent that many fields are missing and entering manually is prone to error.

A more scalable and secure strategy is to deploy Raspberry Pi microcomputers next to each X-ray machine. These low cost, versatile devices can capture essential metadata such as exposure parameters, timestamps, machine logs, and scanning protocols then automatically push it to a GitHub repository. Once a stable automation pattern is established for roughly 5-10 representative machines (covering the main scanner families), the same scripts can be adapted for other devices, minimizing the overhead when integrating new systems.

2. Raspberry Pi as a Metadata Capture and Transfer Hub

2.1 Automated Detection and Parsing

The Raspberry Pi functions as a bridge between the X-ray machine and a remote repository. It continuously monitors the workstation’s output folders, parsing newly generated image files or logs to extract relevant technical details. This information exposure times, resolution, sample identifiers, etc. is packaged into standardized formats (JSON, YAML, or XML).

2.2 Uniformity Across Scanner Brands

Although scanners differ in their data export protocols, a small library of parsing scripts can handle the majority of brands, thanks to shared metadata fields (e.g., file name, resolution, voltage). Once these scripts are validated on a handful of diverse machines (e.g., Nikon Metrology XT H 225 ST, GE phoenix v|tome|x, Bruker SkyScan), minor tweaks typically make them suitable for others.

2.3 Secure Transfer to GitHub

The Pi can authenticate using private keys or tokens, ensuring secure commits to GitHub. Each new metadata file is version controlled immediately, creating a time stamped, immutable record of the scan event. Researchers thus gain confidence that every piece of information about an X-ray session is captured and preserved without manual intervention.

2.4 AI-Enhanced Metadata Analysis Workflows

One of the most powerful features of our GitHub integration is the automated AI analysis pipeline, which consists of two primary workflows:

Workflow #1: MorphoSource Updates Detection

This workflow automatically monitors and reports new additions to the MorphoSource database. For example, a recent detection (Release 2025-01-27_20-44-28) identified:

Workflow #2: CT to Text Analysis

Following the detection of new records, a second workflow performs detailed analysis of the CT data, enhancing the metadata with taxonomic context and anatomical descriptions. This workflow (Analysis 2025-01-27_20-44-48) generates comprehensive descriptions that include:

• Common Names: Automatically adds common names (e.g., “northern madtom” for Noturus placidus) to make the data more accessible to non-specialists
• Anatomical Features: Identifies and describes key structural elements visible in the scan
• Research Context: Highlights potential applications and research value of the scan

Workflow #3: Advanced CT Analysis Pipeline

The third workflow in our pipeline focuses on detailed analysis of CT data, processing both 2D slices and 3D meshes from MorphoSource uploads. This automated analysis provides researchers with immediate insights into specimen characteristics.

2D Slice Analysis Example

The slice analysis workflow examines:

• Slice Characteristics: Progression of detail through different layers
• Structure Analysis: Identification of morphological features
• Orientation Changes: Comprehensive view analysis across multiple angles
• Detail Evolution: Documentation of feature changes across the slice series

3D Mesh Analysis Example

The 3D analysis workflow examines:

Structural Characteristics

• Morphology: Analysis of specimen shape, dimensions, and adaptational features
• Surface Detail: Documentation of texture and anatomical structures

Material Composition

• Density Variations: Analysis of material density through color variation
• Compositional Indicators: Identification of potential mineral content

Feature Analysis

• Anomaly Detection: Identification of unusual features or preservation patterns
• Comparative Views: Analysis across multiple orientations:
  • Default (Y+ Up)
  • Upside Down (Y- Up)
  • Forward 90° (Z- Up)
  • Back 90° (Z+ Up)

This comprehensive analysis pipeline ensures that each new CT upload is automatically processed and analyzed, providing researchers with immediate insights into specimen characteristics. The workflow’s modularity allows for continuous improvement of analysis capabilities and integration of new analytical techniques as they become available.

3. GitHub Branching and Facility Specific Workflows

3.1 Branches for Each Facility or Machine

Given the substantial variety in scanner types and institutional workflows, branching in GitHub emerges as a natural solution. Each lab, hospital, or imaging center can maintain its own branch (e.g., facilityA/metadata, facilityB/metadata).

3.2 GitHub Pages and Templates for Onboarding

In addition to branching, GitHub Pages can host simplified front end documentation to guide new facilities through the setup process. Predefined templates (for metadata structure or Pi scripts) help onboard teams quickly.

Recent Updates and Attestations

Our latest monthly collection (Release 2025-01-26) includes verified attestations, confirming how the data was collected and which Github serverless resources performed the action, that can be viewed here.

3.3 Release Based Tagging

When a facility finalizes a batch of scans or completes a major data ingestion milestone, they can publish a GitHub release. This step provides a static version of the metadata at a particular time, critical for citations and regulatory audits.

4. Centralized Analysis and AI Driven Error Handling

4.1 Main Branch for Consolidated Analytics

While individual branches serve local needs, the main branch can exist as an aggregate repository. Here, data from all facilities is merged (often after automated checks) to form a comprehensive dataset.

4.2 Workflow Failures and Automated Issue Creation

Automation inevitably encounters edge cases incorrect file formats, missing fields, or unresponsive scanners. Rather than silently failing, a robust GitHub Actions setup will generate new issues, document errors, and tag them for immediate review.

5. Claude Assisted Code Generation and Repository Maintenance

5.1 Automated Branch Creation for Fixes

When an error or feature request is detected, AI driven workflows can analyze the issue, create a new branch for fixes, and generate pull requests with proposed changes as seen below using Claude Sonnet to generate code and combine it based on the issue description with Github Actions, see commit here.

5.2 Automated Feature Requests

Beyond error handling, a similar process applies to new features. When someone opens an issue, Claude can parse the request, generate scaffolding code, and place it in a new branch for acceptance testing. This will make the repository more accessible and maintainable by contributors without the need for software development experience.

5.3 Continuous Learning and Refinement

As Claude’s code contributions are reviewed by human experts, the feedback loop informs future suggestions. Over time, the system learns from accepted solutions, refining its approach to both error detection and code generation.