Evaluation of Image Noise from Computed Tomography in an Organ-Specific Anthropomorphic Phantom
summary
The gist
The goal of this study was to utilize an anthropomorphic phantom to: (1) evaluate image noise against the prescribed NI across different organs…
In short
The study used an anthropomorphic phantom to test image noise against prescribed noise indices (NI) across different organs using ASiR-V and DLIR reconstructions. Results showed DLIR reduces noise compared to ASiR-V, with the maximum dose reduction occurring in the liver at a specific NI level. The relationship between noise, dose index, and reconstruction technique is complex and organ-dependent.
Key concepts
- Image Noise (NM)
- This metric measures the deviation or randomness in an image subtraction process across two repeated scans. It quantifies how much random variation exists in the reconstructed image pixels. A lower NM indicates a cleaner, less noisy image.
- Prescribed NI
- The Prescribed Noise Index (NI) is a target level of noise that researchers aim to achieve or control during CT scanning. The study examined how different reconstruction methods perform when aiming for various NI targets across different body parts.
- DLIR vs. ASiR-V
- These are two different image reconstruction algorithms tested on the same CT data. DLIR (Deep Learning Image Reconstruction) was compared to ASiR-V (Adaptive Statistical Iterative Reconstruction). The goal was to see if DLIR provides superior noise reduction capabilities for achieving a target NI.
- CTDIvol
- CTDIvol is a measure of the radiation dose delivered by the CT scanner, specifically volume-averaged. The study investigated how using DLIR reconstruction at equivalent noise levels impacts this dose index, aiming to find maximum possible dose reductions.
Terminology used across episodes
This episode discusses
- Evaluation of Image Noise from Computed Tomography in an Organ-Specific Anthropomorphic Phantom · Paper Radio
The paper
Evaluation of Image Noise from Computed Tomography in an Organ-Specific Anthropomorphic Phantom · Read on arXiv
Department of Radiology, Boston Children’s Hospital · Harvard Medical School · §GE HealthCare
Computed tomography (CT) is used ubiquitously for diagnostic imaging, with novel iterative reconstructions such as hybrid iterative reconstruction with partial model-based component (ASiR-V) and deep learning image reconstruction (DLIR) increasingly utilized. In conjunction, CT vendors have developed a noise index (NI) to reduce radiation dose during protocol development. However, it is unclear how these reconstruction algorithms reduce noise as NI changes, particularly across multiple organs. The goal of this study was to utilize an anthropomorphic phantom to evaluate image noise against the prescribed NI across different organs, determine if DLIR reduces image noise compared to ASiR-V, and use this relationship to determine the maximum volume CT dose index (CTDIvol) reduction at equivalent noise levels. Repeat images were acquired on a GE HealthCare Revolution Apex CT scanner across NI values in an anthropomorphic adult-sized phantom from Kyoto Kagaku with ASiR-V and DLIR. Organs-of-interest were contoured using TotalSegmentator in 3D Slicer. The correlation between the CTDIvol and NI was determined. The standard deviation of the difference in the repeat images was used as a noise metric (NM), correlated with NI, used to determine the NI at equivalent NM between ASiR-V and DLIR, then applied to determine the dose reduction and difference in NM. The relationship between NI and the resulting CTDIvol was found to be proportional to a power law of exponent-2.23. The maximum potential absolute dose reduction from using DLIR was found in the liver at 1.3 mGy with an NI of 7.9. Across all organs, the relative dose reduction increased, the NM-to-NI ratio decreased, and the noise reduction increased across NI. The relationship between image noise and the prescribed NI differed between ASiR-V and DLIR and varied by organ, with model-based dose reductions of up to 1.3 mGy in the liver.
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: "Evaluation of Image Noise from Computed Tomography in an Organ-Specific Anthropomorphic Phantom".
Mira: Here's the gist of the paper. The goal of this study was to utilize an anthropomorphic phantom to: (1) evaluate image noise against the prescribed NI across different organs, (2) determine if DLIR reduces image noise compared to ASiR-V, and (3) use this relationship to determine the maximum volume CT dose index (CTDIvol) reduction at equivalent noise levels Study Purpose The primary objective of the research was to employ an anthropomorphic phantom to achieve three specific aims: (1) evaluate image noise against the prescribed NI across different organs, (2) determine if DLIR reduces image noise compared to ASiR-V, and (3) use this relationship to determine the maximum potential dose reduction at equivalent noise levels The study involved acquiring repeat images on a GE HealthCare Revolution Apex CT scanner across various NI values in an anthropomorphic adult-sized phantom from Kyoto Kagaku using both ASiR-V and DLIR reconstructions Organs of interest were contoured using TotalSegmentator in 3D Slicer The noise metric (NM) was defined as the standard deviation of the subtraction image across the two repetitions divided by the square root of 2 to account for noise propagation The NM-to-NI ratio was defined as the computed NM divided by the prescribed NI Key Findings on Noise and Dose The relationship between NI and CTDIvol was found to be proportional to a power law of exponent-2.23 The NM was dependent on NI using a power law with exponent under 1 with smaller exponents for DLIR The maximum potential absolute dose reduction from using DLIR was found in the liver at 1.3 mGy (approximately 22% of the corresponding ASiR-V CTDIvol) with an NI of 7.9 Across all organs, the relative dose reduction increased, and the noise reduction increased across NI Organ-Specific Dose Reductions The resulting impacts on dose as measured using the CTDIvol can be seen in Figure 7 The dose reduction possible when using DLIR reconstruction to achieve the equivalent NM as the ASiR-V reconstruction is maximized in the liver with a reduction of 1.3 mGy and minimized in the spinal cord with a reduction of 0.2 mGy The optimal NI for the gallbladder is 12.7 and 15.0 for the spinal cord Relationship Between Reconstruction Techniques The relationship between the noise reduction ratio and the prescribed NI is shown in Figure 8 All relationships were positive, meaning that increasing the prescribed NI increases the noise reduction capabilities of the DLIR reconstruction compared to the ASiR-V reconstruction The steepest slope was seen in the liver at 1.02 while the shallowest was seen in the gallbladder at 0.69 indicating their susceptibility to noise reduction as the prescribed NI increases Limitations and Future Directions There were several limitations to this study, including scanning only the inferior portion of the chest, entire abdomen, and superior portion of the pelvis in the phantom The study also only investigated a single slice thickness and a pitch of 1.375 Future work should investigate how these results fit into the broader scope of the diagnostic imaging practice and more in-depth analysis into the noise profiles of each organ using metrics like the noise power spectrum This study explored the effects of varying acquired CT dose levels, in terms of the NI, on the reconstructed image noise, in terms of the NM, in an anthropomorphic phantom with realistic organ volumes and respective CT numbers using the 50% ASiR-V and DLIR reconstruction algorithms Several observations were observed: (1) sharper edges were seen in the noise image at the lowest NI setting in both the ASiR-V and DLIR images, (2) CTDIvol related to NI by a power law with an exponent of-2.23, (3) the ratio of the maximum to minimum mA falls monotonically from 1.53 at a prescribed NI of 5 to 1.18 at a prescribed NI of 30, (4) the relationship between the computed NM and the prescribed NI is not linear, (5) the DLIR reconstruction provided additional noise suppression at the equivalent NI compared to ASiR-V, (6) dose reduction when using DLIR compared to ASiR-V is organ-specific, and (7) the NM-to-NI ratio was found to decrease as the prescribed NI increased for both the ASiR-V and DLIR images References 1.
Kai: Kopka L, Funke M, Breiter N, Hermann KP, Vosshenrich R, Grabbe E.
An anatomically adapted variation of the tube current in CT. [Kai: Studies on radiation dosage reduction and image quality].
Title and authors: Kai: We’ve seen that they are using an anthropomorphic adult-sized phantom from Kyoto Kagaku to evaluate image noise against a prescribed NI across different organs, which sets up their investigation into how these reconstruction algorithms perform in practice.
Mira: And the core of the paper is determining if DLIR reduces image noise when compared to ASiR-V, and then using that relationship to figure out the maximum possible volume CT dose index reduction at equivalent noise levels.
Lev: That leads us straight into their purpose: they want to find a direct link between how much noise is present and the prescribed NI setting, and then see how DLIR interacts with that relationship across different tissues.
Kai: It’s about mapping out this relationship across organs, which is interesting because we know dose reduction isn't uniform everywhere.
Mira: They also mention they contoured the organs of interest using TotalSegmentator in three dee Slicer, which means they are being specific about where they are measuring these noise differences—liver, spinal cord, gallbladder, and so on <ref:2611.07357#pg1>.
The paper's summary: Kai: So, to summarize what the study found is that there’s a clear relationship between the CT dose index volume and the prescribed NI setting; specifically, it follows a power law with an exponent of negative two point two three.
Mira: And they also found that image noise, which they define as the standard deviation of the subtraction image divided by the square root of two to account for noise propagation, is dependent on that prescribed NI using a power law with an exponent under one for DLIR, which is smaller than what’s seen for ASiR-V.
Lev: That’s interesting because if DLIR has a smaller exponent, it suggests that the AI reconstruction method might handle the noise suppression scaling differently depending on the specific organ you are looking at.
Kai: They also found that the maximum potential absolute dose reduction from using DLIR was found to be one point three mGy in the liver when they were at an NI of seven point nine which is roughly twenty-two percent of what ASiR-V achieved at that same NI.
Mira: They also noted that across all organs, both the relative dose reduction and the noise reduction increase as the prescribed NI goes up, but the way these two things scale with NI is not linear.
The paper's improvements: Kai: The paper suggests a few ways this work could be improved. For example, they point out that while DLIR provides additional noise suppression at the equivalent NI compared to ASiR-V, the benefit of using DLIR actually increases as the prescribed NI for ASiR-V reconstruction goes up.
Mira: And they also highlight that there are specific optimal NI values for certain organs; for instance, they found that twelve point seven and fifteen point zero are the optimal NI settings for the gallbladder and spinal cord, respectively.
Lev: From a hardware standpoint, those specific NI suggestions are really useful because it tells us exactly where to dial in our reconstruction parameters to get the best noise performance for a particular anatomy.
Kai: They also suggest that if you want to get even deeper into this, you should look at how the NM-to-NI ratio changes with the prescribed NI, and they find that this ratio actually decreases as the prescribed NI increases for both ASiR-V and DLIR images.
Mira: That relationship between the computed noise metric and the prescribed NI isn't linear, which means we can't just assume a simple straight line to predict performance everywhere.
Conclusion: Kai: So, to wrap up on this study of "Evaluation of Image Noise from Computed Tomography in an Organ-Specific Anthropomorphic Phantom," the main implication is that noise reduction isn't uniform across the body when adjusting NI settings.
Mira: The study confirms that DLIR does provide extra noise suppression compared to ASiR-V at the same NI, but it’s highly dependent on the organ, with the liver showing a bigger dose reduction benefit than the spinal cord in this specific test.
Lev: For us building real hardware, those organ-specific dose reductions mean we need tailored approaches for setting our reconstruction parameters based on what's being scanned.
Kai: It’s important to remember that while they found these relationships, they did have limitations; specifically, the study only scanned the inferior portion of the chest, the entire abdomen, and a superior portion of the pelvis in this phantom.
Mira: And they also had to keep things simple by only investigating a single slice thickness and a pitch of one point three hundred seventy-five during their testing.
Lev: That’s where we can take this work next; we need to see how these relationships hold up when we move into more complex scenarios with variable slice thickness or different acquisition parameters.
Kai: That sounds like the direction for future research, moving from the phantom setup to a broader clinical context, which is where the real challenge lies.
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