Evaluation of quantitative accuracy in 4D DCE‐bCT: Insights from a perfusion phantom study

Abstract

Background

Four‐dimensional dynamic contrast‐enhanced breast CT (4D DCE‐bCT) provides submillimeter spatial resolution and temporal sampling within seconds, offering potential for improved perfusion assessment in breast cancer. Its quantitative accuracy, however, must be validated.

Purpose

To evaluate the accuracy of dynamic iodine concentration quantification in 4D DCE‐bCT using a perfusion phantom with in‐line optical spectroscopy and ultraviolet‐visible (UV/VIS) spectrophotometry as reference standards.

Methods

A dynamic flow phantom with a 3 mm internal diameter was used to approximate tumor‐feeding vessels. Potassium iodide (KI) contrast was introduced using block‐pulse injections at 1, 3, and 5 mg I/mL and a gradual inflow‐outflow protocol to reproduce bolus passage. In‐line optical spectroscopy and UV/VIS spectrophotometry provided reference measurements. Dynamic bCT acquisitions were performed using a previously developed sparse‐view protocol. A static pre‐contrast scan was reconstructed with the iterative maximum‐likelihood polychromatic algorithm for CT (IMPACT), and post‐contrast dynamic scans were reconstructed using the prior image constrained compressed sensing (PICCS) algorithm. Acquisitions were performed at 65 kV with 0.25 mm Cu filtration, 80 mA for the pre‐contrast scan (300 projections over 10 s) and 32 mA for the post‐contrast scans (444 projections per 100 s), using a 5 ms pulse width. Projection data were corrected for scatter and motion before reconstruction, and the static prior was then subtracted from post‐contrast volumes. Object‐size effects were evaluated using the 3 mm tube and 5, 10, 15, and 20 mm hollow spheres filled with non‐flowing iodine at 1, 3, and 5 mg I/mL. The tube interior was segmented using thresholding and morphological cleanup, while spherical phantoms were segmented by identifying the sphere center and extracting all enclosed voxels. Voxel values were converted to iodine concentration using a pre‐determined calibration curve, and agreement with reference standards was evaluated using peak concentrations and time‐concentration profiles, with errors reported as absolute and relative differences.

Results

In dynamic block‐pulse experiments, peak iodine concentrations measured by 4D DCE‐bCT were systematically underestimated relative to ground truth, with relative errors of approximately 12% to 39%. In gradual inflow‐outflow experiments, the peak iodine concentration derived from 4D DCE‐bCT was 5.24 ± 0.61 mg I/mL, whereas optical ground truth measured 5.17 mg I/mL. Static experiments demonstrated an average underestimation of 0.23 ± 0.14 mg I/mL across target sizes and concentrations and no relationship between iodine estimation and target size.

Conclusion

4D DCE‐bCT enabled acquisition of time‐varying iodine concentration profiles in a 3 mm tube phantom. Gradual changes were captured more accurately compared to rapid changes, consistent with temporal averaging. These results demonstrate the feasibility of dynamic iodine imaging with 4D DCE‐bCT while highlighting current limitations and motivating further evaluation in anatomically realistic phantoms and pilot clinical studies.

1. INTRODUCTION

Breast cancer is the most diagnosed cancer and the leading cause of cancer‐related deaths among women globally. In 2024, approximately 2.4 million new cases were diagnosed, and around 693 660 deaths were reported. ¹ This burden is expected to grow, with both incidence and mortality projected to rise by 2040. ²

Breast cancer heterogeneity influences cancer progression, treatment outcomes, and prognosis. ³ MRI is currently the gold standard for disease staging and treatment planning, especially in patients undergoing neoadjuvant chemotherapy, since it helps assess treatment response and detect residual disease. ⁴ , ⁵ , ⁶ , ⁷ However, MRI has several limitations, including difficulty in detecting calcified structures and limited spatial and temporal resolution for tracking contrast uptake. Additionally, the lack of a standardized measure for tissue characteristics across vendors hampers the consistency of its quantitative analysis. ⁸

To address these limitations, Prionas et al. enhanced lesion detection using contrast‐enhanced dedicated breast CT (CE‐bCT), ⁹ while Caballo et al. demonstrated the feasibility of 4D dynamic CE‐bCT (4D DCE‐bCT) for high‐resolution functional imaging. ¹⁰ Building on these advancements, a bCT system was optimized for this dynamic contrast‐enhanced scanning, with the specific aim of achieving accurate iodine quantification at high spatio‐temporal resolution, supported by an x‐ray spectrum optimized to enhance iodine concentration quantification,11 and new reconstruction algorithms.12 4D DCE‐bCT aims to use iodine quantification to capture the full profile of breast tumors, including heterogeneous regions, for staging, treatment planning, and monitoring.

The aim of this study is to experimentally evaluate the quantitative accuracy of dynamic iodine imaging in 4D DCE‐bCT using a functional physical phantom, and to assess the influence of object size on quantification accuracy. Ground truth iodine concentrations were obtained with in‐line optical spectroscopy, supported by ultraviolet‐visible (UV/VIS) spectrophotometry of collected samples. The simplified phantom minimizes unnecessary complexity while replicating key imaging characteristics, providing a controlled setting to evaluate both dynamic iodine quantification and object‐size effects across the complete 4D DCE‐bCT pipeline.

2. MATERIAL AND METHODS

A dedicated breast CT (bCT) system was used to perform 4D dynamic contrast‐enhanced imaging of a custom‐made breast perfusion phantom. Projection data were corrected for scatter and motion and reconstructed with algorithms optimized for iodine quantification. Dynamic acquisitions were performed on the 3 mm diameter tube. Additional static acquisitions were performed with non‐flowing iodine in the 3 mm tube and 5, 10, 15, and 20 mm hollow spheres to evaluate object‐size effects.

Iodine concentrations derived from the images were validated against in‐line optical spectroscopy, the input concentrations, and UV/VIS spectrophotometry of collected samples. The experimental and analysis workflow is illustrated in Figure 1.

FIGURE 1.

Workflow of the experimental setup and analysis for the 4D DCE‐bCT phantom study. The process starts with phantom preparation and pump protocol configuration, providing the input for the generation of a theoretical curve, followed by two parallel acquisition paths: In‐line optical spectroscopy with UV/VIS analysis (left) and 4D bCT imaging with scatter and motion correction, reconstruction, and VOI segmentation (right). Dynamic acquisitions were performed on the 3 mm diameter tube. Static acquisitions used the same imaging protocol but with non‐flowing iodine in the 3 mm tube and 5, 10, 15, and 20 mm hollow spheres, to assess object‐size effects. Outputs from both paths were combined for quantitative accuracy assessment of iodine concentration measurements. Source: This figure was created in BioRender. https://BioRender.com/3p1n5nr.

2.1. Breast CT system and acquisition parameters

A modified dedicated bCT system (Vera, Koning Corp., Norcross, Georgia, USA) was used for this study. Tube current and exposure were set according to previous work, ¹¹ , ¹² to optimize iodine quantification. Pre‐contrast scans were performed using 0.25 mm Cu filtration at 65 kV, 80 mA, and 5 ms pulses.

Post‐contrast scans used 65 kV, 32 mA, and 5 ms pulses. The pre‐contrast sequence comprised 300 projections over a single revolution in 10 s, while post‐contrast imaging captured 444 projections over 10 revolutions in 100 s for each of the three dynamic phases, covering both wash‐in and wash‐out, for a total of 1332 projections. There is an inherent 5 s gap between consecutive dynamic phases due to system exposure reset and preparation.

2.2. 4D DCE‐bCT acquisition and reconstruction

The acquired 4D DCE‐bCT projection images were first corrected for scatter using a previously developed Monte Carlo‐based method, then corrected for rigid motion, and finally reconstructed with the iterative maximum‐likelihood polychromatic algorithm for CT (IMPACT) and prior image constrained compressed sensing (PICCS) algorithms. ¹² , ¹³ , ¹⁴ , ¹⁵

For scatter correction, a validated Monte Carlo simulation, previously described and updated to include the breast phantom, ¹⁶ was performed to replicate the geometry and acquisition settings of the bCT system. ¹⁷ A single scan was reconstructed and segmented as input into the Monte Carlo simulation to obtain primary and scatter projection estimates. The simulation used a 65 kV x‐ray spectrum with 0.25 mm Cu filtration and a 1st HVL of 4.47 mm Al, ¹¹ tracked 10⁹ photons, and generated 300 projections at 1.2° intervals. The resulting simulated scatter projections were scaled to the appropriate magnitude and subtracted from the measured projections.

The motion correction algorithm performs a rigid translation on the projection data, determined using a phase correlation method, to align the projections with a reference image. ¹⁵ Although motion would not normally be expected in a phantom experiment, this correction was necessary because the static experiments required replacing the pre‐contrast water with iodine. This exchange introduced small shifts when the lid was removed and repositioned to inspect for air bubbles, which could otherwise introduce misalignment artifacts in the reconstructed images. As the motion‐correction step was incorporated into the processing pipeline, all acquisitions were processed using this pre‐processing step.

The IMPACT algorithm extends the classical maximum‐likelihood transmission reconstruction (MLTR) approach by modeling x‐ray attenuation as the sum of photoelectric and Compton scattering components while only allowing the presence of a predefined number of material combinations as constraint. ¹³

The IMPACT reconstruction assumes an effective energy of 35 keV; all attenuation values reported in this study are given at 35 keV. The reference energy of 35 keV was chosen in the development stage where it matched the average energy of the x‐ray spectrum of the previous breast CT scanner. The reference energy was kept at 35 keV for the newer hardware even though the average energy now is closer to 45 keV to keep results and evaluations consistent. This reference energy is effectively a scaling factor similar to the selected energy for a virtual monochromatic image created from a spectral CT scan, and does not influence the reconstruction nor the quantification of iodine (in mg I/mL) other than setting the scale for the numerical values it generates.

PICCS reconstruction ¹⁴ was then applied to the scatter‐corrected data, using the IMPACT image as a prior. Each time frame was reconstructed from 44 evenly spaced projections spanning 10 s, with the time assigned to each frame corresponding to the average time of acquisition of its contributing projections. ¹² For each 444‐projection phase, this yielded 19 reconstructed time points, each separated by 5 s, with a 10 s temporal window, thus partially overlapping with neighboring time points. In between post‐contrast phases, the 5 s inter‐sequence gap described above results in approximately 17.5 s between the last time point of sequence N and the first time point of sequence N + 1. The reconstructed images had an isotropic voxel spacing of 0.25 mm.

In terms of dose, the static prior acquisition (300 projections at 65 kV, 80 mA, 0.25 mm Cu, 5 ms pulse length delivered 10.2 mGy air kerma at isocenter. Each dynamic phase (444 projections at 65 kV, 32 mA, 0.25 mm Cu, 5 ms pulse length) delivered 6.2 mGy. Using a normalized glandular dose coefficient (DgN) of 0.762 mGy/mGy based on the phantom dimensions, the total mean glandular dose for a full dynamic acquisition was 21.9 mGy. ¹²

2.3. Iodine contrast concentration range in breast and tumor tissue

To test the quantitative accuracy of dynamic iodine imaging in 4D DCE‐bCT, an appropriate iodine concentration range must be selected. Previous phantom studies in quantitative contrast‐enhanced breast x‐ray imaging used iodine concentrations of 0–2 mg I/mL, 0–3 mg I/mL, and 0–5 mg I/mL. ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³

Clinical CT studies have reported mean iodine concentrations in breast tumors between 1 and 3 mg I/mL, with maxima up to 6 mg I/mL, although values depend strongly on contrast protocol and acquisition phase. Volterrani et al. ²⁴ used Iopamidol 370 mg I/mL at 1.5 mL/kg (late arterial phase) and reported a mean of 2.7 ± 1.0 mg I/mL (maximum 5.6 mg I/mL), identifying 1.7 mg I/mL as an optimal diagnostic cutoff. Hasse et al. ²⁵ applied Iohexol 350 mg I/mL at 1.43 mL/kg (venous phase) and observed 1.9 ± 0.6 mg I/mL with maxima of 6 mg I/mL, while Chen et al. ²⁶ used Ioversol 320 mg I/mL at 1.2 mL/kg (arterial and venous phases) and reported mean values of 1.1 mg I/mL (maximum 1.38 mg I/mL). Wilpert et al. ²⁷ used Iopromide 370 mg I/mL at 1.8 mL/kg, and image acquisition began 85 s after the start of injection. Based on these findings, a maximum iodine concentration of 6 mg I/mL was selected to ensure clinical relevance, since this value encompasses the reported lesion concentrations.

2.4. Phantom design

The simplified breast perfusion phantom is modelled after a standard bCT quality control phantom with a maximum diameter of 13 cm (Koning Corp, Norcross, Georgia, USA). To simulate the x‐ray attenuation of breast skin, the wall of the phantom is 1.5 mm thick and printed from clear resin material using a 3L SLA printer (Formlabs Inc., Sommerville, Massachusetts, USA). ²⁸ The phantom is filled with olive oil to mimic fatty tissue. ²⁹ , ³⁰ The breast‐shaped phantom has a total height of 125 mm and a maximal internal diameter of 120 mm, and it is closed with a 3D‐printed acrylic lid (Figure 2). The outer contour of the phantom has a perimeter of 300 mm. To avoid tumor‐like shapes (e.g., a sphere) acting as chambers and altering the fluid dynamics, a U‐shaped inlet tube with an inner diameter of 3 mm and a height of 80 mm is integrated into the lid, with a path length corresponding to a 200 mm perimeter.

FIGURE 2.

Design of the breast perfusion phantom. The schematic illustrations (left and center) show the phantom structure and positioning of components, while the right image displays the actual phantom filled with olive oil to mimic fatty tissue.

2.5. Phantom pumping system

The phantom was connected to a dedicated pumping system comprising three programmable syringe pumps (AL‐4000, World Precision Instruments, Hertfordshire, UK): One for contrast input, one for water input, and one for fluid extraction. The system was prefilled with water, and flexible medical‐grade tubing (5 mm outer diameter, 3 mm inner diameter, 70 cm length) was used among the parts to ensure consistent flow and minimal resistance.

2.6. Optical system for monitoring the iodine concentration in the contrast medium

An in‐line optical spectroscopy system, illustrated in Figure 3 and previously described in detail, was used to measure the iodine concentrations at the phantom inlet (P1) and outlet (P2) during the dynamic protocols.³¹The setup contains a fiber‐coupled LED (MINTF4, Thorlabs, Newton, New Jersey, USA) and collimators (F950, Thorlabs) for sample illumination and a photodetector (PDA10A2 Si Fixed Gain 200–1100 nm, Thorlabs) and oscilloscope (Picoscope 7, PicoTechnology, Cambridgeshire, UK) to record the signal. In our previous study, the setup was able to measure iodine concentrations in the range of 0–6 mg I/mL with an accuracy of 0.008 (± 0.07) mg I/mL and with < 0.1% precision. The measurement cell was modified from earlier prototypes to improve dynamic performance (shown in Appendix A, Figure A1). The updated measurement piece, with a smaller cross‐section compared to the rectangular piece, minimized mixing artifacts and allowed for a more accurate measurement of time‐dependent iodine concentrations.

FIGURE 3.

Schematic of the phantom pumping system (described in Section 2.5) for dynamic contrast delivery. Photodetectors (P1 and P2) monitor the flow (as described in Section 2.6), and 4D DCE‐bCT x‐ray imaging captures contrast changes. The sample extraction pump can extract samples at pre‐defined time‐points. Source: This figure was created in BioRender. https:// BioRender.com/99nu46l.

Potassium iodide (KI) was selected as the contrast medium for its optical detectability and comparable x‐ray attenuation to clinical agents. Adjusting for molecular weight differences enables the preparation of KI solutions with iodine concentrations (mg I/mL) equivalent to those observed in patient studies, making it a suitable substitute for phantom studies. ³¹

The optical setup comprised a green LED coupled to a fiber optic cable and collimator to produce a parallel light beam through the measurement cell. A second collimator on the opposite side directed the transmitted light into a fiber optic cable connected to a photodetector.

Before dynamic measurements, calibration curves were obtained by recording transmission values for eight known KI concentrations (0, 0.5, and 1–6 mg I/mL in 1 mg I/mL steps) at both P1 and P2. The concentration and transmission were related by a one‐phase decay function, defined as:

$$Y=Plateau+\left(Y_0-Plateau\right)·\exp \left(-K·X\right)$$

Function fitting was performed using Prism software (GraphPad Software, Boston, MA, United States) and 95% prediction bands were added. These functions were applied during the experiments to convert measured light intensities into iodine concentrations. ³¹

2.7. Dynamic measurements

Two types of measurements were conducted to evaluate the quantitative performance of 4D DCE‐bCT under different enhancement patterns. The first used discrete, well‐defined iodine pulses, allowing direct validation of system accuracy since the exact iodine concentration for each pulse was known. The second used a gradual inflow‐outflow protocol designed to mimic physiologically relevant perfusion curves and assess the system's ability to track continuous, gradual changes in iodine concentration over time.

2.7.1. Block pulse protocol

Three block‐pulse protocols were acquired, each comprising a 20 s iodine pulse at 24 mL/min, preceded and followed by water to flush the system. Fixed pulse concentrations of 1, 3, and 5 mg I/mL were used. The iodine dilutions were prepared using potassium iodine (7.6 mg I/mL) and water, using an analytical scale with a readability and repeatability of 0.1 mg (Entris II Essential, Sartorius, Göttingen, Germany).

For each protocol, a single 4D dynamic breast CT acquisition was performed, consisting of a 10 s pre‐contrast phase followed by a single 100 s post‐contrast phase, as described in Section 2.2. The iodine pulse was initiated during the post‐contrast phase and was fully contained within this acquisition, allowing complete temporal sampling of the pulse within one dynamic sequence.

2.7.2. Gradual inflow‐outflow protocol

To generate a controlled, gradually increasing and decreasing iodine curve, a 24 mL mixing container was added to the pumping setup. In this chamber, the contrast and water were mixed before entering the phantom. The pumping program was set to deliver a contrast input of 7.6 mg I/mL at a rate of 24 mL/min for 100 s, followed by 200 s of water input at the same rate. This produced a peak intensity of approximately 6 mg I/mL within the first 100 s, followed by a decline to near 0 mg I/mL over the subsequent 200 s. The experiment was then repeated twice without 4D DCE‐bCT acquisition to allow for fluid sampling and to avoid perturbations in the flow that could compromise the in‐line optical measurements.

Using 3‐way valves connected to syringes positioned before the inlet and after the outlet measurement sections, 1 mL samples were collected at the peak input time (140 s) and peak output time (160 s) for subsequent analysis (see Section 2.8). The acquisition timing for the gradual inflow‐outflow protocol is shown in Figure 4.

FIGURE 4.

Time sequence for the gradual inflow‐outflow protocol, consisting of a 10 s pre‐contrast scan followed by three consecutive 100 s post‐contrast scans (444 projections each) separated by 5 s gaps. Contrast injection began 20 s after the start of the first post‐contrast phase, with wash‐in occurring during this phase, reaching a peak during the second phase and wash‐out extending into the subsequent phases. Source: This figure was created in BioRender. https://BioRender.com/ vmp8hmj.

2.7.3. Theoretical curves

The variation of iodine concentration in the tubing over time can be estimated from the pump settings and system volumes by solving an ordinary differential equation (ODE). ²⁸ The change in concentration inside the tube is calculated using a theoretical approximation as explained in Appendix C.

2.8. Fluid sample extraction and analyses

Samples obtained during the gradual inflow‐outflow measurements were analyzed using a UV/VIS spectrophotometer (UV‐2600, Shimadzu, Kyoto, Japan) with an absorbance accuracy of 0.002 (equivalent to 6 µg I/mL). Before sample analysis, a calibration curve was established by measuring the absorbance of eight KI solutions with known iodine concentrations (0, 0.5, and 1–6 mg I/mL in 1 mg I/mL steps). A fitted curve was then used to relate absorbance values to iodine concentration. This relationship was applied to the measured absorbance values of the collected samples to determine their iodine concentrations.

2.9. Static acquisitions to evaluate object‐size effects

Static scans were performed to determine how object size influences iodine quantification. The 3 mm tube and hollow spheres of 5, 10, 15, and 20 mm diameter, fabricated using the same 3D‐printed process and resin material as the main phantom components, ³⁰ were each filled with non‐flowing iodine solutions at 1, 3, and 5 mg I/mL (Figure 5). For each object size, a single scan was acquired to generate the pre‐contrast image, followed by a post‐contrast acquisition for the tube and the spheres filled with 1, 3, and 5 mg I/mL. All images were reconstructed using the IMPACT algorithm.

FIGURE 5.

Schematic of the static phantom configurations used to evaluate object‐size effects on iodine quantification.

Voxel values from the post‐contrast scans were converted to iodine concentration using the calibration procedure described in Section 2.10, following the quantitative pipeline as described in 2.11.2.

2.10. X‐ray calibration for iodine quantification

Calibration samples were prepared by diluting potassium iodide (KI) stock solution with water to obtain concentrations ranging from 0 to 6 mg I/mL (0, 0.5, and 1–6 mg I/mL in 1 mg I/mL steps). Each dilution was stored in 50 mL tubes (28 mm inner diameter) and placed in a 3D‐printed holder inside the breast phantom filled with olive oil. ¹² All samples were scanned and reconstructed as described in Section 2.2 for static prior acquisitions. A water‐filled tube (0 mg I/mL) was first acquired and designated as the reference prior, which was subtracted from the images with iodine dilutions to obtain iodine‐only images and compute the calibration curve.

2.11. Pre‐processing and segmentation

2.11.1. Dynamic tube experiments

The pre‐contrast reconstruction served as a prior. It was subtracted voxel‐wise from each post‐contrast volume, and the resulting difference images were cropped to a fixed volume of interest. A binary mask was generated from the prior by slice‐wise Gaussian smoothing (σ = 0.5) and fixed thresholding at 0.26 cm⁻¹. This threshold was selected from the histogram of the prior image, allowing retention of the tubing and lumen while excluding lower‐attenuation background structures.

After thresholding, 3D morphological cleanup was applied: connected components smaller than 250 voxels were removed and internal voids up to 500 voxels were filled. The mask was then eroded in 3D using a spherical structuring element (radius = 10 voxels) to exclude the 3D printed tube, ensuring that only the iodine‐containing region remained in the segmented volume. The eroded binary mask and the masked prior volume were saved for subsequent analysis. The resulting volume encompasses the masked and eroded 20 mm‐long tube with a 3 mm diameter. At a fluid flow rate of 24 mL/min, this corresponds to 3.5 s of fluid flow. Combined with the 10 s temporal window from image reconstruction, this yields a total temporal window of 13.5 s per 4D DCE‐bCT data point in the dynamic tube experiments.

2.11.2. Static sphere experiments

For the static sphere acquisitions, a spherical VOI was defined in Python in the pre‐contrast and post‐contrast images. The objects were segmented using Otsu thresholding from the Skimage.filters library to generate a binary mask. A 3D Euclidean distance transform was then applied to the masked volume, and the voxel with the maximum distance to the object boundary was identified as the center of the sphere. Using this center coordinate, a spherical VOI with a 1 mm offset to the actual border was generated. Voxel values contained within the ROI were extracted for quantitative analyses.

2.12. Quantitative analysis

For the static experiments, the mean and standard deviation values within the mask were computed. The mean of the pre‐contrast image was subtracted from the post‐contrast images, and the standard deviation was computed by:

$$\sigma =\sqrt{{\sigma }_{pre}^2+{\sigma }_{post}^2}$$

These measurements were conducted to investigate the effect of object size on iodine quantification. The mean and standard deviation were calculated separately for the scans acquired before and after refilling the spheres with iodine solutions. This separate analysis was performed to account for any positional shifts that may have occurred during the refilling procedure.

To investigate the relationship between voxel size and measured x‐ray attenuation, the 1 mg I/mL, 3 mg I/mL, and 5 mg I/mL tube experiment was reconstructed using different voxel sizes. Subsequently, the mean and standard deviation values within the VOI were analyzed and added to Figure D.1 in the Appendix. To investigate the partial volume effects near the borders of the different targets, the signal was plotted as a function of distance from the center in Figure D.2 in the Appendix.

For the dynamic experiments, quantitative analysis was performed by applying the eroded mask to each cropped, subtracted dataset. Voxel values within the mask were extracted as raw attenuation and converted to iodine concentration using the calibration factor derived from KI calibration samples. Unlike the static analysis, which was based on summary statistics within each sphere, the dynamic scan did not involve manual interventions, and thus it was possible to perform the analysis using voxel‐wise quantification, matching the patient studies.

2.13. Comparison with reference data

For both the block‐pulse and gradual inflow‐outflow measurements, optical signals were recorded at the P1 and P2 positions, then time shifted based on the tube length between P1 and P2 and the breast phantom, and averaged to obtain a single optical reference curve. The optical signal was averaged over a 13.5 s temporal window to be similar to the 4D DCE‐bCT datapoints.

For the block pulses (1, 3, and 5 mg I/mL), concentration‐time curves were generated and the 4D DCE‐bCT‐derived concentration‐time curves were compared with the with the optical measurements. The agreement between 4D DCE‐bCT and ground truth was assessed by comparing peak concentrations and reporting absolute and percentage errors.

For the gradual inflow‐outflow sequence, UV/VIS samples were drawn at the same two positions near the peak, analyzed by spectrophotometry, and averaged to yield a single UV/VIS reference value. The 4D DCE‐bCT‐derived concentration‐time curve was compared with the averaged optical measurement, averaged UV/VIS samples, and a 13.5 s time‐averaged theoretical curve. Comparisons included peak concentration, wash‐in and wash‐out dynamics, and qualitative agreement of curve shapes. The mean and standard deviation of the differences across time between the optical signal and the 4D DCE‐bCT signals were computed for the three post‐contrast phases. A linear regression analysis was done to assess the relationship between the optical and 4D DCE‐bCT signal, and a Bland‐Altman plot was generated.

3. RESULTS

3.1. In‐line spectroscopy and UV/VIS calibration curves

The calibration curves with 95% prediction bands for the two in‐line spectroscopy systems are shown in Figure B1 in Appendix B. At the phantom entrance (P1), the fitted equation is:

$$y=0.087+\left(1.004-0.087\right)exp\left(-0.644X\right)$$

with where $X$ is the iodine concentration (mg I/mL) and $y$ is the measured transmission (I/I₀). The root‐mean‐square error (RMSE) is 0.005.

At the phantom exit (P2), the relationship is defined as:

$$y=0.071+\left(1.002-0.071\right)exp\left(-0.639X\right)$$

with an RMSE of 0.004.

The calibration curve for the tested iodine concentrations in the UV/VIS spectrophotometer is illustrated in Figure B2 of Appendix B. The relationship between iodine concentration and absorbance at 525 nm is defined as:

$$y=0.32X+0.02$$

with an R ² value of 0.99. The concentrations of the samples were estimated using this relationship, and the results are added as data points in the figure.

3.2. X‐ray calibration results

The x‐ray calibration curve (Figure 6) demonstrated a linear relationship, with a slope of 41.08 mg I/mL per cm⁻¹ and a standard error (Sy.x) of 0.09 mg I/mL, indicating close agreement between measured data and the fitted line.

FIGURE 6.

Calibration curve relating iodine concentration to the measured linear attenuation in the subtracted iodine‐only images. Blue markers indicate measured values. Error bars represent the standard deviation of repeated measurements. The solid blue line shows the linear regression fit, grey dotted lines represent the 95% confidence interval of the fit. The residual standard error (Sy.x = 0.09 mg I/mL) quantifies the scatter of measured values around the fitted regression line.

3.3. Dynamic acquisitions: Optical measurements versus 4D‐BCT

3.3.1. Block pulse measurements

Block pulse time‐concentration curves for iodine dilutions of 1, 3, and 5 mg I/mL are shown in Figure 7. The corresponding peak concentrations and errors are summarized in Table 1. The 4D bCT results show wash‐in and wash‐out at the expected times. The 3 mg I/mL and 5 mg I/mL measurements show similar underestimation, with percentage errors of 14.5% and 11.8%, whereas 1 mg I/mL shows a higher percentage error of 37.4%.

FIGURE 7.

Block pulse measurements obtained with 4D DCE‐bCT and optical time‐ averaged results for 1, 3, and 5 mg I/mL. The vertical error bars represent the standard deviation within the single 4D DCE‐bCT measurement. The horizontal error bars represent the temporal window of the breast CT measurement.

TABLE 1.

Peak concentration comparison between 4D DCE‐bCT and ground truth for block pulse measurements.

Nominal concentration (mg I/mL)	Optical peak ± std (mg I/mL)	4D DCE‐bCT peak ± std (mg I/mL)	Absolute error (mg I/mL)	Percentage error (%)
1	0.99 (0.96–1.00)	0.62 (0.21–1.03)	0.37	37.4
3	2.56 (2.51–2.60)	2.19 (1.64–2.74)	0.37	14.5
5	4.34 (4.19–4.51)	3.83 (3.21–4.45)	0.51	11.8

3.3.2. Gradual inflow and outflow

The top figure in Figure 8 shows the gradual inflow‐outflow sequence. The 4D DCE‐bCT‐derived concentration‐time curve is plotted alongside the theoretical profile, the averaged optical signal, and the averaged UV/VIS measurement. The UV/VIS measurement (6.06 mg I/mL) coincided with the peak of the theoretical curve, while the optical average peaked at 5.17 ± 0.27 mg I/mL. The 4D DCE‐bCT peak concentration was 5.24 ± 0.61 mg I/mL, overlaps with the optical signal (1.3% difference), and has a 10.6% underestimation compared to the theoretical curve. Table 2 shows the average and std. dev. of the differences between the optical signal and breast CT signal for all data points (all phases), and for post‐contrast phases 1, 2, and 3. Overall, the 4D bCT results show similar dynamics to the ground truth. The two bottom figures in Figure 8 show the linear regression analyses and the Bland‐Altman plot of the optical signal versus the 4D DCE‐bCT signal. The relationship between the two datasets can be described by Y = 0.99X‐0.41, with an R ² value of 0.98. The confidence interval for the y‐intercept was −0.52 to −0.30 mg I/mL; since this interval does not include zero, the optical and 4D DCE‐bCT data were statistically unequal. The Bland‐Altman plot shows a bias of 0.41 mg I/mL between the optical and 4D DCE‐bCT signals.

FIGURE 8.

Top: Comparison of 4D DCE‐bCT‐derived concentration‐time curve (pink) with averaged optical reference (green) and averaged ground‐truth input concentration (spectrophotometry, dark blue) during the gradual inflow‐outflow sequence. The vertical error bars represent the standard deviation within the single 4D DCE‐bCT measurement. The horizontal error bars represent the temporal window of the breast CT measurement. The time‐averaged theoretical concentration profile is shown in light blue for reference. Bottom left: the linear regression analysis of the optical signal vs the 4D DCE‐bCT signal; the confidence interval of the y‐intercept was −0.52–−0.30 mg I/mL. Bottom right: The Bland‐Altman plot of the optical and 4D DCE‐bCT signal, showing a bias of 0.41 mg I/mL.

TABLE 2.

Mean and std. dev. of the differences between the optical and breast CT signal for the entire 4D DCE‐bCT scan and each separate post‐contrast phase.

Phase	Mean difference (mg I/mL)	Standard deviation of the difference (mg I/mL)
All	0.43	0.47
1	0.51	0.51
2	0.41	0.56
3	0.31	0.34

3.4. Evaluation of object‐size effects

Figure 9 shows bCT images of the 10‐mm sphere at different concentrations. Figure 10 shows the estimated iodine concentration as a function of object size for nominal concentrations of 1, 3, and 5 mg I/mL. Data points represent mean measured concentrations, with error bars indicating the standard deviation of voxel values within the VOI. The numerical labels correspond to the mean values. The average underestimation was 0.23 (± 0.14) mg I/mL. No trend was found between object size and iodine concentration estimation. Partial‐volume effects occur over approximately 2 voxels for all targets, as shown in Appendix D, but were avoided by staying 1 mm from the border during quantitative analysis of the VOIs.

FIGURE 9.

Cropped breast CT images of the 10 mm spheres filled with different iodine concentrations and the VOIs used for quantitative analyses. Window: 0.164, level: 0.266.

FIGURE 10.

The quantified iodine concentrations as a function of object size for nominal concentrations of 5 mg I/mL (dark blue triangles), 3 mg I/mL (green squares), and 1 mg I/mL (light blue circles). Error bars indicate standard deviation.

4. DISCUSSION

This study experimentally evaluated the quantitative accuracy of dynamic iodine imaging in 4D DCE‐bCT using a simplified perfusion phantom, an in‐line optical reference, ground‐truth input concentrations, and extracted samples as the reference standards. The results showed that the reconstruction and calibration pipeline consistently captured time‐varying iodine concentration profiles.

The x‐ray calibration curve derived from KI dilutions was highly linear over the investigated range. However, at low attenuation values, small absolute errors resulted in substantial relative concentration errors. For example, at 1 mg I/mL, a measured attenuation of 0.022 cm⁻¹ corresponded to 0.84 mg I/mL, representing a 16% underestimation.

The tube with a 3 mm inner‐diameter used in this work was intended to approximate feeding vessels that develop during tumor angiogenesis, providing a deliberately challenging but clinically relevant test.

The average underestimation was 0.23 (± 0.14) mg I/mL in the static measurements and ranged from 0.37 mg I/mL to 0.51 mg I/mL in the dynamic measurements. Previous studies evaluating iodine quantification accuracy in CT using various dual‐energy approaches reported mean errors ranging from 0.24 (± 0.007) mg I/mL to 0.44 (± 0.24) mg I/mL for iodine concentrations between 0.2 mg I/mL and 4 mg I/mL. ³² In addition, a study of photon‐counting breast CT reported a root mean square error of 0.53 mg I/mL for iodine concentrations between 2.5 mg I/mL and 25 mg I/mL. ³³ Therefore, overall, the iodine quantification errors observed in our study are consistent with those reported in literature.

In the dynamic block‐pulse experiments, peak iodine concentrations were measurable but underestimated relative to the ground‐truth input concentrations by approximately 12%–37% across the investigated concentration range. Interestingly, the underestimation observed in the gradual inflow‐outflow experiment was substantially smaller (∼16%). This difference is likely related to the temporal characteristics of the acquisitions. The block‐pulse experiments consisted of relatively short‐duration concentration changes (∼20 s), while each reconstructed 5 s time point represented an effective temporal averaging window of approximately 13.5 s due to the reconstruction process and spatial averaging within the VOI. Consequently, the rapid concentration transitions in the block‐pulse experiments were temporally blurred, leading to attenuation of the measured peak concentrations. In contrast, the more gradual concentration changes in the inflow‐outflow experiment could be followed more accurately by the breast CT system, resulting in reduced underestimation. Linear regression analysis between the optical and 4D DCE‐bCT measurements, however, still showed a statistical difference between the two signals, and the Bland‐Altman plot of the full dynamic signal showed a bias of 0.41 mg I/mL.

In patient imaging, contrast enhancement kinetics are generally expected to evolve more gradually than the rapid transitions investigated in the block‐pulse experiments. Typically, the initial wash‐in phase occurs over approximately 100 s, followed by delayed enhancement behavior characterized as persistent enhancement (> 10% increase after wash‐in), wash‐out (> 10% decrease after wash‐in), or plateau enhancement patterns, as commonly described in dynamic breast MRI. ³⁴ Based on the results of the present phantom experiments, particularly the gradual inflow‐outflow measurements, such clinically relevant enhancement dynamics are expected to be captured well, though with a bias, with 4D DCE‐bCT.

For the 1 mg I/mL measurements, a relatively large underestimation was observed across both the static and block‐pulse experiments, which cannot be fully explained by the approximately 16% calibration bias at low iodine concentrations alone. In addition to this calibration‐related underestimation, the low signal‐to‐noise ratio at 1 mg I/mL increased the sensitivity of the measurements to image noise. Furthermore, the temporal characteristics of the block pulse experiments might have further increased the underestimation. Together, these factors likely contributed to the larger underestimations observed compared with the higher iodine concentrations, although the absolute underestimation was similar to that in the higher‐concentration experiments.

The gradual inflow–outflow sequence was compared with the theoretical curve, UV/VIS samples, and optical measurements, whereas the block pulse experiments were only compared with the optical measurements. We chose this approach because no mixer was used in the block pulse experiments, and a near‐ideal block pulse was achieved only when a small air bubble was present between the syringe and the inlet tubing, effectively preventing mixing between the fluids at the inlet side. In the 3 mg I/mL and 5 mg I/mL experiments, however, some degree of mixing was observed. Therefore, we considered the experimentally measured optical signal to be the most appropriate reference for comparison with the 4D bCT measurements, as it accurately reflected the actual concentration profile delivered to the phantom rather than the intended block pulse.

The in‐line optical measurements showed better agreement with the nominal input concentration values for lower iodine concentrations. At 5 mg I/mL, larger deviations were observed, indicating increased error of the optical reference at higher iodine concentrations. This behavior is consistent with reduced discrimination of higher absorbance levels by the photodetector at increased iodine concentrations, as reflected by the wider confidence intervals in the one‐phase decay function at higher iodine concentrations, as shown in Figure B.1 in the Appendix. Although the updated 3D‐printed optical measurement piece reduced mixing artifacts and improved quantification of time‐dependent iodine concentrations, it did cause this declined transmission accuracy at iodine concentrations above 5 mg I/mL, requiring discrete UV/VIS sampling at peak times. Moreover, although the UV/VIS spectrophotometer was highly accurate, manual sampling at fixed time points remained prone to minor timing errors, reflected as horizontal error bars in the gradual inflow‐outflow graph.

This study presents the initial characterization of the quantitative accuracy of 4D DCE‐bCT. Although the results clearly indicate the capabilities of this novel imaging technique, more comprehensive characterization of its accuracy and precision is needed in future studies. Most importantly, the dynamic experiments should be repeated multiple times on multiple days to assess the imaging system's repeatability and reproducibility. Furthermore, the location dependence of the quantitative accuracy and precision should be investigated, and the effect of different breast sizes should be studied. Moreover, the impact of breast tissue heterogeneity, for example, varying degrees and locations of glandular and adipose tissue in the background, should be studied using more sophisticated heterogeneous anthropomorphic phantoms.

A limitation of the current study is the relatively high noise level observed in individual dynamic 4D bCT measurements, as reflected by the standard deviations within the measured VOIs. Although this variability limits quantitative interpretation of isolated time points, dynamic imaging primarily relies on the evaluation of temporal enhancement patterns across the complete time series rather than on single measurements alone. In the present experiments, 57 consecutive time points were acquired, enabling characterization of the temporal iodine dynamics despite the noise present in individual frames. Furthermore, the phantom data were intentionally reconstructed with only pre‐processing for scatter, excluding any denoising, in order to evaluate the baseline performance of the acquisition and reconstruction framework. A 4D similarity filter could improve the quantitative results substantially. ³⁵ , ³⁶ However, optimization of such a method is highly application‐dependent and may be more appropriately tuned in future in vivo studies, where anatomical structures differ substantially from those in the homogeneous phantom experiments, making any filtering results more challenging but realistic.

Our simplified phantom design allowed us to validate the end‐to‐end pipeline without interference from breast tissue heterogeneity, though at the cost of not reproducing perfusion complexity or microvascular networks. More complex phantoms will provide additional insight, but the next step will be pilot clinical imaging studies to assess feasibility in patients and to refine reconstruction parameters for clinical application.

4D DCE‐bCT is intended for applications in breast cancer staging, treatment planning, and treatment response monitoring. Compared with MRI, the technique offers higher combined spatial and temporal resolution, as well as direct quantitative iodine concentration measurements. Because the imaging field is confined to the breast, repeated acquisitions can be performed while minimizing radiation exposure to the remainder of the body.17 This may enable repeated quantitative assessment of tumor perfusion dynamics at the sub‐tumor level, including voxel‐wise mapping of parameters such as wash‐in time, peak enhancement, and mean transit time. In DCE‐MRI and spectral CT, such TIC parameters have been associated with malignancy risk as well as histopathological prognostic and predictive biomarkers. ²⁶ , ²⁷ , ³⁴ , ³⁶ , ³⁷ , ³⁸ Consequently, 4D DCE‐bCT may facilitate identification of intratumoral heterogeneity and support more personalized treatment strategies.

5. CONCLUSION

This study demonstrated that 4D DCE‐bCT enables acquisition of time‐varying iodine concentration profiles in a 3 mm tube using in‐line optical, ground‐truth input concentrations, and UV/VIS reference measurements. Gradual changes were better captured compared to rapid changes, consistent with temporal averaging. These findings support progression toward more anatomically realistic phantoms and initial clinical evaluation.

CONFLICT OF INTEREST STATEMENT

Ioannis Sechopoulos is a Scientific Advisory Board member of Koning Corp.

Supporting information

Supporting information

Goris LC, Pautasso JJ, Michielsen K, Sechopoulos I. Evaluation of quantitative accuracy in 4D DCE‐bCT: Insights from a perfusion phantom study. Med Phys. 2026;53:e70719. 10.1002/mp.70719