Inline Detection of 0.3 mm Fish Bones Using Low-Energy X-ray TDI Imaging – Why Sub-Millimeter inspection Is a Platform Stability Challenge — Not a Resolution Specification

Detecting 0.3 mm fish bones inline at industrial conveyor speeds represents a boundary condition problem in food inspection X-ray. At this scale, contrast is limited, photon statistics become critical and mechanical stability directly impacts detectability.

Below we demonstrate detection feasibility using:

Individual slice images and low/high-energy TDI reconstructions. These were evaluated to determine signal formation, SNR limits and achievable detection performance.

Detecting sub mm fish bones inline is often framed as a detector resolution problem.

It is not.

Inspection set up

Source Parameters

  • Tube Voltage: 50 keV
  • Tube Current: 4.5 mA
  • Source Distance: ~800 mm

Acquisition Parameters

  • Conveyor Speed: 0.5 m/s
  • Detector Pitch: 1 mm × 1 mm
  • Integration Time: 2000 µs
  • 8 TDI rows
  • Gain: 9.375 pC
Using magnifcation effect to see 0.3mm object resolution with a 1mm pixel
 

To use geometric magnification so that a 1 mm detector pixel “looks like” 0.3 mm at the object, you need the effective object-plane pixel:

  • SID = source-to-detector distance

  • SOD = source-to-object distance

  • ODD = object-to-detector distance

If your SID is 800 mm the object would need to be ~240 mm from the source and ~560 mm from the detector.

 

At this scale, detection becomes a system-level signal integrity challenge governed by:

  • Photon statistics
  • Signal chain stability
  • Mechanical precision
  • Calibration repeatability
  • Scatter control

OEMs who approach this as a pixel-size specification exercise typically encounter unstable performance in production.

 

 

The Physics Reality

Fish bone is not metal.

It is a low-Z, calcium-based structure embedded in soft tissue with only marginal attenuation contrast at 50 keV.

At 0.3 mm thickness, the expected differential attenuation between bone and surrounding muscle produces only a ~1–2% signal variation per pixel.

That means:

The system must operate with effective SNR levels above ~50–80 to reliably separate bone from noise in real production conditions.

This is not a nominal imaging task but is a precision signal detection task.

At line speeds of  0.5 m/s, motion during one integration equals exactly one pixel (1 mm). TDI accumulation across 8 rows improves SNR by ~√8 ≈ 2.8×.

At 0.3 mm thickness, attenuation contrast in soft tissue is small. Detection is therefore limited by:

  • Quantum noise
  • Scatter contamination
  • Detector stability
  • Motion blur
  • Electronic noise floor

The task becomes a signal-to-noise engineering problem, not a nominal resolution problem.

 

Strategic Takeaway

At sub-millimeter detection limits:

Photon counting is not a silver bullet.

Detection margin is primarily governed by:

  • Photon statistics
  • Stability
  • Scatter management
  • Calibration precision

Photon counting shifts noise composition — it does not change the fundamental √N limit.

If the platform is already quantum limited, architecture choice becomes a trade-off between:

  • Flux handling
  • Stability
  • Cost
  • Complexity
  • Scalability

 

A single slice is insufficient for stable 0.3 mm detection.

TDI Reconstruction

After 8-row accumulation:

  • Noise is reduced.
  • Bone contrast becomes perceptible.
  • Stability improves significantly.

However:

The detection margin remains narrow.
Any instability directly impacts detectability.

 

Detector Gain Setting (9.375 pC)

High gain improves sensitivity but risks:

  • Reduced dynamic range
  • Saturation in thick regions
  • Amplified electronic noise if frontend not optimized

Detection of 0.3 mm bones requires:

  • Low electronic noise floor
  • Stable charge integration
  • Precise gain calibration across rows

 

Individual Slices

  • Slice-level images operate near the quantum noise floor.
  • Row-dependent gain variations are visible.

Structured noise and scatter are present.

 

 

Detection is quantum-noise dominated.

Photon counting would:

  • Remove electronic noise floor
  • Offer modest scatter suppression
  • Introduce count-rate management complexity

Expected improvement in 0.3 mm detectability:

Likely 10–25% margin increase — not 2×.

Increasing TDI rows from 8 to 16 would produce similar or greater improvement with lower system risk.

Conclusion:

  • Low energy acquisition is required for sub-millimeter bone detection.

Inline detection of 0.3 mm fish bones at:

  • 50 keV
  • 0.5 m/s
  • 1 mm pitch
  • 8-row TDI

is technically feasible but operates at the edge of SNR limits.

Sub-millimeter bone detection is not a component problem.
It is a platform integrity problem.

At 50 keV with 8-row TDI:

  • The physics allows detection.
  • The margin is narrow.
  • Stability determines commercial success.

The real differentiator is not resolution.

It is whether the system can deliver the same SNR — every shift, every day, every installation.

For more information, contact 

Paul Hurtado, Head of Sales and Marketing at Sens-Tech.