Efficiency limits in in-line imaging
summary
The gist
In-line phase-contrast imaging can deliver significant gains in image quality compared to attenuation-based imaging at the same radiation dose The ultimate limit of this gain is determined by the…
In short
In-line phase-contrast imaging can significantly improve image quality over attenuation-based methods. Combining known phase retrieval techniques with Point Spread Function (PSF) deconvolution, especially using Contrast Transfer Function correction, maximizes this gain. The ultimate limit of this quality improvement is determined by the ratio of the real decrement to the imaginary part of the sample's complex refractive index.
Key concepts
- Phase-Contrast Imaging (PBI)
- This technique uses X-ray phase shifts caused by sample variations to create images, offering better contrast than traditional attenuation methods. The quality gain in PBI is directly related to how well the phase information is recovered from the measured signal.
- Point Spread Function (PSF) Deconvolution
- PSF deconvolution sharpens images by mathematically reversing the blurring effects introduced by the imaging system's optics and detector. Combining this with phase retrieval methods allows for a significant boost in spatial resolution and image quality.
- Contrast Transfer Function (CTF) Correction
- The CTF correction method is a specific technique used to maximize image quality gain when combining phase retrieval and PSF deconvolution. This combination achieves the highest theoretical gain, which is limited by the sample's physical properties.
- Relative Interaction Strength ($\gamma$)
- This parameter represents the relative strength of X-ray absorption versus refraction channels in PBI. It is a key determinant of imaging quality; its ratio dictates the ultimate upper limit for how much image quality can be improved.
Terminology used across episodes
This episode discusses
The paper
Efficiency limits in in-line imaging · Read on arXiv
School of Physics, University of Melbourne · School of Physics and Astronomy, Monash University
In-line phase-contrast imaging, also known as propagation-based imaging or Gabor holography, is capable of delivering significant gains in image quality, compared to attenuation-based imaging at the same radiation dose. Image quality metrics, including signal-to-noise ratio and spatial resolution, are considered, and their relationship to the amount of Shannon information in the image is discussed. It is demonstrated that it is possible to increase the image quality gain by combining known phase retrieval methods with deconvolution of the point-spread function (PSF) of the imaging system, with the highest gain achieved using the Contrast Transfer Function correction method in combination with the PSF deconvolution. The ultimate limit of this gain is determined by the ratio of the real decrement to the imaginary part of the complex refractive index of the imaged sample. Analytical results for the image quality limits in in-line imaging are obtained using a Gaussian model for the detector PSF. A numerically simulated example confirming the theoretical results is also presented.
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: We're looking at the paper: "Efficiency limits in in-line imaging".
Mira: Here's the gist of the paper. In-line phase-contrast imaging can deliver significant gains in image quality compared to attenuation-based imaging at the same radiation dose The ultimate limit of this gain is determined by the ratio of the real decrement to the imaginary part of the complex refractive index of the imaged sample Image Quality Metrics and Limits Image quality metrics, including signal-to-noise ratio and spatial resolution, are considered, and their relationship to the amount of Shannon information in the image is discussed It is demonstrated that it is possible to increase the image quality gain by combining known phase retrieval methods with deconvolution of the point-spread function (PSF) of the imaging system, with the highest gain achieved using the Contrast Transfer Function correction method in combination with the PSF deconvolution The ultimate limit of this gain is determined by the ratio of the real decrement to the imaginary part of the complex refractive index of the imaged sample Phase Retrieval and Gain Factors The corresponding gain in imaging quality can be defined as the ratio of SNR in (i) The transport-of-intensity equation (TIE) is a continuity equation for paraxial monochromatic scalar light beams (Teague, 1983).
Kai: It expresses the rate of change of intensity with respect to propagation distance along the optical axis, in terms of the intensity and phase of the beam over planes perpendicular to that axis This result can be understood in terms of the underlying imaging physics The relative interaction strength of the X-ray absorption and refraction channels can be expressed as a ratio of the corresponding scattering cross-sections, which is equal to gamma, making this parameter a key determinant of imaging quality in PBI Role of PSF Deconvolution The fact that the spatial resolution in the image plane is improved after a coherent free-space propagation by a factor proportional to 1/NF gamma rather than 1/2 NF gamma as claimed in (Gureyev et al., 2017) The gain factor G2D quantifies the gain in the image SNR after coherent free-space propagation and TIE-Hom retrieval, in comparison with the “contact” (ABI) image in the plane z = 0 When combined with PSF deconvolution, this gain factor can reach gamma, i.e., it can indeed reach the predicted ultimate upper limit of efficiency Comparison of Reconstruction Methods The CTF-Hom+PSF-deconvolution approach can deliver superior performance to the TIE-Hom+PSF-deconvolution, in terms of achievable SNR, spatial resolution and the intrinsic imaging quality, QS The gain factor 1/2 Q
PBI: /Q
ABI: S D S D,2,2 = gamma, which is achieved in eq.(30) The most practically relevant case for biomedical imaging is represented by the situation where only a single Fresnel fringe is detectable in PBI images In this case, the CTF-Hom+PSF-deconvolution image reconstruction can still potentially deliver an advantage, compared to the TIE-Hom+PSF-deconvolution method The gain factor 1/2 Q
PBI: /Q
ABI: S D S D,2,2 = gamma, which is achieved in eq.(30) depends only on the relative strengths of the phase (refraction) and absorption scattering channels Numerical Validation The results of the simulated phase retrieval are shown in Fig.4 The top row (panes (a)-(b)) contains the images obtained by the TIE-Hom and CTF-Hom retrieval from the propagated image shown in Fig.3(c), while the bottom row contains images obtained using the same methods with the deconvolved propagated image from Fig.3(d) The reconstructed images in Figs.4(c-d), obtained using the combinations of the phase retrieval and PSF deconvolution, are visually sharper (have higher spatial resolution) compared to the corresponding images in Figs.4(a-b) obtained without the deconvolution The line profile through the edge of the image in Fig.4(d), obtained using a combination of CTF-Hom retrieval with the deconvolution of the PSF (purple long-dash line in Fig.5), almost perfectly reproduces the original edge profile in the “ground truth” image in Fig.3(a) (solid black line in Fig.5) Future Directions In the future, it may be interesting to investigate the use of QS as a loss metric in training of deep learning (AI) models for tasks such as image deconvolution and denoising This option may be particularly beneficial in view of the known low sensitivity of some of the traditionally used loss metrics, such as peak signal-to-noise (PSNR), SSIM and least-squares difference, with a ground truth image, to variations in the propagation phase contrast and phase retrieval Acknowledgements The authors acknowledge funding from the Australian National Health and Medical Research Council (grant No.
Mira: APP2011204) The main results obtained in Sections 5, 7 and 8 are summarized in Table 3 This indicates, in particular, that the CTF-Hom+PSF-deconvolution approach can deliver superior performance to the TIE-Hom+PSF-deconvolution, in terms of achievable SNR, spatial resolution and the intrinsic imaging quality, QS The ultimate upper limit of the gain factor G3D is determined by eq.(5) and is asymptotically equal to 1/2 gamma - when ln 1 gamma > 1 In both methods, the single determinant of the gain in image quality was the parameter 1/2 gamma, unlike the case of the original TIE-Hom method where the gain factor was equal to 1/2 NF gamma and thus also depended on other parameters such as the detector resolution, the propagation distance and the X-ray wavelength The dependence of the image quality gain on the spatial resolution of the detector can be eliminated by means of deconvolution of the detector PSF, and (ii) The ultimate upper limit of is determined solely by the relative strengths of the phase (refraction) and absorption scattering channels Appendix on Validity A sufficient condition for validity of eq.(4) is 2, 1/2 gamma > obj max rho lambda R or 2 Delta > obj R lambda The most practically relevant case for biomedical imaging is represented by the situation where only a single Fresnel fringe is detectable in PBI images The most practically relevant case for biomedical imaging is represented by the situation where only a single Fresnel fringe is detectable in PBI images The most practically relevant case for biomedical imaging is represented by the situation where only a single Fresnel fringe is detectable in PBI images The most practically relevant case for biomedical imaging is represented by the situation where only a single Fresnel fringe is detectable in PBI images The most practically relevant case for biomedical imaging is represented by the situation where only a single Fresnel fringe is detectable in PBI images Numerical Simulation Details A numerically simulated homogeneous sample had an approximate range of transmission values (0.99, 1.0) and the ratio of the real decrement to the imaginary part of refractive index gamma = 100 The detector was assumed to have perfect quantum efficiency, eta = 1, square pixels with size 1.25 x 1.25 μm squared and a circular Gaussian PSF with sigma det = 12.5 μm, i.e., with spatial resolution Delta det / sqrt(4)43μm The registered “flat-field” photon fluence (without the object in the beam) was assumed to be Poisson distributed with a mean value of 1111 photons per detector pixel, which corresponded to noise with the standard deviation of 3% in the flat-field corrected image Table Summary The results of the simulated phase retrieval are shown in Fig.4 The SNR, spatial resolution and intrinsic image quality values measured in the simulated images are generally also in good agreement with the theoretical predictions Table 4 contains a summary of the above results, including a restatement in terms of the two-dimensional “intrinsic imaging quality characteristic” (Gureyev et al., 2014, 2016) The quantitative results in Table 4 are consistent with the qualitative observations discussed above in relation to Figs.
Title and authors: Kai: So we're moving into the title and authors part of this paper, "Efficiency limits in in-line imaging."
Mira: The authors are T. E. Gureyev, D. M. Paganin, and H. M. Quiney from the School of Physics at the University of Melbourne and Monash University respectively.
Lev: I'm not really familiar with their specific setups, but I can tell you they’re working on X-ray phase-contrast imaging here in line with what we've been discussing about PBI and PB-CT.
Kai: They are focusing this work primarily on X-ray PBI of biological samples, which is where a lot of the real application for this kind of technique lies right now.
Mira: They are also noting that their methods and results are broadly applicable, even though they focus on biological samples.
Lev: That’s interesting because often when you build these things for real hardware, we have to deal with very specific detector geometries or source characteristics.
Kai: Exactly, and the paper addresses how to optimize both the hardware setup—the X-ray source and the detector—and the software for image processing.
Mira: They frame it as two interdependent tasks: optimizing the imaging hardware setup depends on your image processing strategy, and vice versa.
Lev: That interdependence is always where things get tricky in practice; you optimize one thing, and it messes up the other.
Kai: So this paper is laying out that framework for how to approach those two optimization tasks together.
The paper's summary: Mira: Now looking at the summary of "Efficiency limits in in-line imaging," it boils down to what they are trying to achieve.
Kai: They are showing that in-line phase contrast imaging can give significant gains over attenuation methods for the same radiation dose, and they are exploring how to measure this image quality gain using metrics like signal-to-noise ratio and spatial resolution.
Lev: So they’re not just claiming a better image; they’s quantifying exactly how much better it can be, which is important for setting expectations.
Mira: They demonstrate that combining known phase retrieval methods with PSF deconvolution can increase the image quality gain, and the highest gain comes from using Contrast Transfer Function correction with PSF deconvolution.
Kai: And this ultimate limit of that gain is tied to that ratio we talked about earlier: the real decrement to the imaginary part of the complex refractive index of the sample.
Lev: So they’re linking their practical image quality gains directly back to fundamental optical physics of how light goes through a material.
Mira: They also discuss how they can reach this limit by using CTF hom plus PSF deconvolution, which they show yields superior performance compared to TIE hom plus PSF deconvolution in terms of signal-to-noise ratio and spatial resolution.
Kai: The main implication is that the combination of these specific tools allows them to achieve a better result than just using TIE hom alone for certain conditions.
The paper's improvements: Lev: What about the actual improvements they suggest? What are they proposing to change in the methodology?
Kai: They propose combining known phase retrieval methods with deconvolution of the point spread function, which we already discussed as a way to increase image quality gain.
Mira: They highlight that achieving the highest gain requires using Contrast Transfer Function correction in combination with PSF deconvolution.
Lev: That means if you want the biggest boost, you have to use both techniques together instead of just one or the other.
Kai: And they also talk about a specific way to handle the Contrast Transfer Function during CTF hom reconstruction by excluding an epsilon vicinity around each critical frequency in the denominator to limit noise amplification.
Mira: That exclusion helps manage noise when you do that, which is something we have seen before in reconstruction algorithms.
Lev: It’s like adding a regularization term that keeps the math stable when it gets too noisy at certain frequencies.
Kai: And they also suggest using the intrinsic imaging quality characteristic QS as a loss metric for training deep learning models instead of things like peak signal-to-noise ratio or SSIM.
Mira: That’s because QS is often better at capturing the variations caused by propagation phase contrast and phase retrieval when you're dealing with those specific effects.
Lev: So they are pushing away from traditional metrics toward something that might actually be more robust for learning what we need to learn in these complex imaging problems.
Conclusion: Kai: So wrapping up the paper, they summarize their main points about efficiency limits in in-line imaging.
Mira: The key takeaway is that the most practically relevant case for biomedical imaging is when you only detect a single Fresnel fringe in PBI images.
Lev: They showed that even in that specific scenario, their CTF hom plus PSF deconvolution reconstruction still has an advantage over the TIE hom plus PSF deconvolution method.
Kai: That advantage is quantified by that gain factor one over two times the ratio of QS for PBI over QS for ABI in the plane z equals zero, which they write as gamma in equation thirty.
Mira: And this gain factor only depends on those relative strengths of the phase and absorption scattering channels.
Lev: So it’s a very clean result when you consider what drives the gain factor down to just those two channel strengths.
Kai: It shows that for this specific case, it's not dependent on detector resolution or propagation distance anymore.
Mira: And they confirm that they eliminated the dependence on detector resolution by using PSF deconvolution in their analysis.
Lev: So if you look at the results of "Efficiency limits in in-line imaging," you see how that combining methods simplifies the dependence to just those scattering channel strengths.
More episodes
- 2611.14082-Two atoms with strong light-induced dipolar interactions
- 2611.14040-Characterization of BNLs 20 and 30 micrometer thick LGADs for the Electron Ion Collider
- 2611.13399-Lift-element-based dynamic network of vortex gust-airfoil interactions
- 2611.13476-Hopfion Superlattices in 3D Twisted Moir'e Light Fields
- 2611.13642-Investigation of the timing performance of double-sided p-type 3D silicon sensors produced by CNM
- 2611.13830-SiPM optical readout of a PICOSEC Micromegas precise timing detector
- 2611.14152-Where are the shortest-path nodes in spatial networks?
- 2611.14219-Sub-Ice: semi-automated mapping of basal channels from ice shelf surface elevation data
- 2611.16341-SUMNO: Spatially Unrolled Multigrid Neural Operator for Fluid Flow Surrogates
- 2611.16090-Multi-UTE Myelin Fraction Mapping: Repeatability and Cross-Field-Strength Reproducibility at 1.5T and 3T