Multi-energy computed tomography and material quantification: Current barriers and opportunities for advancement

Abstract

Computed tomography (CT) technology has rapidly evolved since its introduction in the 1970s. It is a highly important diagnostic tool for clinicians as demonstrated by the significant increase in utilization over several decades. However, much of the effort to develop and advance CT applications has been focused on improving visual sensitivity and reducing radiation dose. In comparison to these areas, improvements in quantitative CT have lagged behind. While this could be a consequence of the technological limitations of conventional CT, advanced dual-energy CT (DECT) and photon-counting detector CT (PCD-CT) offer new opportunities for quantitation. Routine use of DECT is becoming more widely available and PCD-CT is rapidly developing. This review covers efforts to address an unmet need for improved quantitative imaging to better characterize disease, identify biomarkers, and evaluate therapeutic response, with an emphasis on multi-energy CT applications. The review will primarily discuss applications that have utilized quantitative metrics using both conventional and DECT, such as bone mineral density measurement, evaluation of renal lesions, and diagnosis of fatty liver disease. Other topics that will be discussed include efforts to improve quantitative CT volumetry and radiomics. Finally, we will address the use of quantitative CT to enhance image-guided techniques for surgery, radiotherapy and interventions and provide unique opportunities for development of new contrast agents.

1. INTRODUCTION

1.A. Organization and scope of the paper

Conventional, dual-energy, and photon-counting detector computed tomography (CT) can each be used to make quantitative measurements that inform and improve clinical decision-making, and each evolution in these technologies has enabled the use of novel image types for this purpose. This paper focuses on the use of quantitative metrics derived from CT for clinical use, with an emphasis on applications of multi-energy CT (MECT). Section 1 gives background information regarding conventional CT, dual-energy CT (DECT), and photon-counting detector CT (PCD-CT). Section 2 details use of DECT for several quantitative applications with an emphasis on the historical use of conventional CT. There is relatively little research performed using dual-energy CT as an input to pharmacokinetic models; rather, DECT has primarily been investigated as a surrogate for perfusion CT due to the lower radiation dose and imaging time. Therefore, perfusion or dynamic contrast-enhanced CT quantitation is beyond the scope of this paper. Comprehensive reviews of perfusion CT are cited here.^1–3 The limitations of quantification with conventional, dual-energy, and PCD-CT are summarized in Section 3, while Section 4 covers emerging developments in the field of quantitative MECT. Discussion of the quantitative MECT’s potential to impact clinical practice is presented in Section 5, with conclusions in Section 6.

1.B. Conventional CT

The first clinical CT scanner was introduced in the early 1970s, and quantitative uses for the system were in development by the end of the decade. Computed tomography data consists of x-ray transmission measurements taken at a large number of angles as an x-ray tube rotates around a patient. The x-ray attenuation is a function of the linear attenuation coefficient (μ, measured in units of cm⁻¹) along the x-ray path as well as the thickness of the material it passes through. With data taken from multiple angles, it is possible to use image reconstruction algorithms, such as filtered backprojection, to calculate a map of the linear attenuation coefficients throughout a cross section of the body. However, the linear attenuation coefficient varies based on the energy spectrum, and therefore visualization is aided by normalizing the raw attenuation data to the linear attenuation coefficient of water (μwater). The CT number, measured in Hounsfield Units (HU) was defined as:

$$HU=1000\times \frac{{\mu }_{\mathit{\text{tissue }}}-{\mu }_{\mathit{\text{water }}}}{{\mu }_{\mathit{\text{water }}}}$$ (1)

where μ represents the linear attenuation coefficient of a given material. The CT number of air, which has virtually no x-ray attenuation, is therefore approximately −1000 HU while the CT number of water is 0 HU regardless of the energy spectrum used for scanning. It became clear that quantitative measurements of CT number could be correlated with disease processes. Early quantitative imaging applications focused primarily on uses of the CT number for tissue and disease characterization,^4–6 bone mineral density measurement,^7–13 and volumetry.^14,15 Standardized CT number thresholds were developed, which allowed for differentiation between materials such as intracranial hemorrhage and calcification^16,17 and to define contrast enhancement in solid renal masses^18–21 with acceptable sensitivity and specificity. As technology has improved, significant advances enabled faster scanning, thinner images, minimization of artifacts, and decreased variability — verified through required quality control programs²² — in quantitative measurements in patients over time.

1.C. Dual-energy CT

Dual-energy CT was conceived by Godfrey Hounsfield in 1973,²³ and the concept was further investigated by Alvarez and Marcovski²⁴ in 1976. Dual-energy CT became commercially viable with the release of the first dual-source DECT system in 2006. In DECT, two x-ray spectra are acquired with different peak energies to provide energy dependent information about the photoelectric and Compton scatter components of a voxel’s attenuation. This enables advanced applications in which material-specific images^24–26 (e.g., iodine) and synthetic monoenergetic images²⁷ may be generated. Material-specific images are generated using material decomposition algorithms. These algorithms utilize a linear combination of the photoelectric and Compton effects for two or more chosen materials within a given voxel to estimate the volume fraction of each basis material present. Since material decomposition algorithms typically assume that only the chosen basis materials are present within each voxel, proper identification of the materials relevant to each application is paramount for accurate quantification. Several examples of clinically available applications for specific material decompositions are shown in Table I. Synthesized monoenergetic images are created by summing the attenuation of the estimated amount of iodine and water in a voxel at a specific energy, typically between 40 and 200 keV. By measuring the attenuation at multiple synthesized energy levels, a simulated transmission spectrum can be generated and potentially used to identify tissue types. Monoenergetic images have been used semi-quantitatively for improved image contrast, thus this review will focus primarily on the quantitative use of material-specific images.

Table I.

Common dual-energy CT material decomposition schemes and their application.

DECT material decompositiona	Applications	Anatomical region	Relevant pathologies
Iodine/water or Soft tissueb	Virtual removal of iodinated contrast	Head and neck, thoracic, abdominal, pelvic	Renal cell carcinoma, pulmonary function, pulmonary embolism, contrast extravasation vs hemorrhage, tumor characterization and staging
Iodine/water or soft tissue/adipose	Virtual removal of iodinated contrast in liver, fat quantification	Abdominal (liver)	Fatty liver disease, hepatocellular carcinoma, liver metastasis
Calcium/water or soft tissue	Virtual calcium/bone removal, calcium quantification	Musculoskeletal, cardiovascular, abdominal, Head and neck	Bone marrow edema, vascular plaque, renal stones, calcification and hemorrhage differentiation
Calcium/hemorrhage/brain	Brain hemorrhage evaluation	Head and neck	Differentiate intracranial hemorrhage and calcification
Uric acid	Renal stone composition gout	Abdominal extremity	Differentiate calcific and uric acid-based renal stones and gout crystais in non-contrast CT exams

^aApplications may not be commercially available from all vendors.

^bThe use of water or soft tissue is vendor-specific.

1.D. Photon-counting detector CT

Recently, PCD-CT has become technologically feasible for true multi-energy CT scanning, and has been used in multiple preclinical and clinical studies.^28–35 This represents an evolution of the methods for collecting CT data within the energy domain; similarly to DECT, the data can be processed into material-specific or virtual monoenergetic images. Photon-counting detectors use direct conversion techniques, where incident x rays are absorbed in semiconductor detector material [e.g., cadmium telluride (CdTe), cadmium zinc telluride (CZT), gallium arsenide (GaAs) or silicon], generating a cloud of electron-holes paired with charges proportional to the energy of the incoming photon. Electrons are then attracted to the readout electronics by means of a large bias voltage placed over the semiconductor layer. In an ideal detector, each x ray absorbed by the semiconductor layer would result in an individual electrical pulse, and the height of the pulse would indicate the x-ray energy. By setting different energy thresholds, incoming photons can be placed in discrete bins, and small pulses caused by electronic noise can be removed from the data. Because the data can theoretically be split into multiple energy bins with minimal overlap, there is potential for PCD-CT to provide opportunities for improved quantification with material decomposition relative to DECT. However, existing PCDs do not operate as ideal detectors since significant spectral overlap remains a concern (see Section 3.C for additional details). Photon-counting detectors do provide temporally and spatially co-registered projections for all energy windows, which can minimize artifacts relative to DECT.³⁶ Furthermore, PCD-CT can allow for imaging a wider variety of potential contrast agents,^35,37 minimized electronic noise relative to DECT acquisitions,³⁸ and concurrent use of multi-energy and high spatial resolution modes³⁶ that is not available on all DECT systems.

A major advantage of PCD-CT over DECT is the potential improvement in material decomposition, particularly for materials with K-edges within the diagnostic energy range. While the differing physical spectra provide some unique material information in DECT, PCD-CT allows users to select specific energy bins to provide improved spectral separation, which may result in better accuracy for material differentiation compared to DECT. Additionally, the number of energy bins and their energy ranges can be chosen to further optimize multi-material decomposition. For example, four-material decomposition is possible in cases where multiple materials with K-edges in the diagnostic range are present³⁹ (see Section 4.A for a discussion of novel contrast agents). This avoids the use of a priori constraints such as volume conservation that are commonly used in DECT material decomposition algorithms. Application of more energy bins with PCD-CT material decomposition could improve material differentiation for disease characterization, evaluation for response to therapy or enable simultaneous administration of multiple contrast agents. In phantom studies, PCD-CT has been shown to accurately measure material concentrations of iodine, gadolinium, and calcium.^40,41

2. CURRENT APPLICATIONS OF QUANTITATIVE CT

2.A. Tumor volumetry

Recent efforts spearheaded by the Radiological Society of North America’s quantitative imaging biomarkers alliance (QIBA) have approved a profile for measurement of tumor volume change⁴² and have reached a consensus regarding CT volume assessment of small lung nodules.⁴³ For measurement of tumor changes in metastatic disease, the longitudinal monitoring of treatment response and disease progression in solid tumors is critical. Quantitative measurements of tumor volume change have gone through multiple standards since the publication of the first World Health Organization guidelines in 1981,⁴⁴ including response evaluation criteria in solid tumors (RECIST), where the longest diameter of multiple lesions are measured and followed over time.⁴⁵ In the most recent update, RECIST 1.1, response to treatment is determined by summing the measured diameters of up to five representative lesions or metastatic lymph nodes that measure at least 10 mm in diameter at baseline and comparing the sum over time.⁴⁶

There are several documented issues with these measurements. RECIST criteria related to diameter measurements has several limitations that may be resolved by using tumor volumetry as a metric, including the variability in tumor shape and the lack of measurements in orthogonal planes. Nevertheless, RECIST 1.1 has been utilized as a primary endpoint in Phase III drug trials and is considered to provide strong evidence for complete or partial response. QIBA has provided guidance about the confidence clinicians can expect under standard acquisition conditions. Overall, they assert that a 95% confidence interval of 30% for change in tumor volume from baseline is achievable based on a conservative estimate of variation in quantitative values due to slight differences in protocol, patient positioning, contrast administration, and measurement variability.⁴² Therefore, a change of at least 30% or more volume should be suggestive of significant change from baseline, and may be indicative of either response to treatment or progressive disease.

One primary weakness of using CT volumetry and RECIST criteria is the response of certain hypervascular tumors on CT imaging. Specifically, the response of tumors such as Gastrointestinal Stromal Tumors (GIST) and certain renal cell carcinomas (RCC) to certain biologic drugs, including imatinib (Gleevec®, Novartis, Basel, Switzerland) and other molecularly-targeted antiangiogenic agents, do not demonstrate volume changes but rather variation in tumor perfusion metrics such as reduced vascularity. In 2007, Choi et al. published a study that indicated that the size of imatinib-treated GIST was not the primary marker of response to treatment; rather, attenuation differences on contrast-enhanced CT correlated strongly with a decrease in standardized uptake value on PET/CT for marked response to treatment.^47,48 The initial Choi criteria for partial response was defined as a decrease in the largest unidimensional diameter of a lesion of >10% or more than a 15% decrease in CT number.^47,48 Similarly, progressive disease was determined by a tumor diameter increasing by more than 10%.^47,48 An update to this rule, known as Modified Choi Criteria, requires both a decrease in diameter and a drop in attenuation to qualify as a partial response.⁴⁹ Therefore, it is critical to appreciate disease heterogeneity, with respect to type of solid tumor involved as well as the therapeutic treatment provided, in order to appropriately apply volumetric CT metrics.

2.B. Bone mineral density

The prevalence of osteoporosis, clinically significant bone mineral density loss, amongst the population of the United States in women above 50 yr of age is estimated to be approximately 29.9% according to a recent study by Wright et al.⁵⁰ For those 80 yr of age or older, the prevalence of significant bone loss increased to 46.3% in men and 77.1% in women.⁵⁰ Fragility fractures related to osteoporosis frequently occur in the hip, vertebrae, wrist, and humerus, resulting in severe morbidity and increased financial burden, particularly among elderly populations.^51,52

Dual-energy x-ray absorptiometry (DEXA) is the primary method for measurement of BMD in use clinically; however, DEXA measurements of areal bone density are affected by patient height,⁵³ weight,^54–56 and distribution of adipose tissue.^57–59 Additionally, there is a risk of misinterpreting DEXA-based BMD measurements in pediatric subjects.⁶⁰ Dual-energy x-ray absorptiometry measurements are also demonstrated to be unreliable in patients with degenerative changes in the bone.^61,62 The tomographic reconstructions in CT have been shown to improve quantitation in many subjects in which DEXA fails,⁷ although QIBA has not yet developed a profile on this topic. Quantitative CT BMD measurements are primarily made in reference to calibration phantoms, which provide known quantities of bone-equivalent materials such as potassium phosphate⁶³ (QCT PRO; Mindways Software, Inc.; Austin, TX) or a combination of water-equivalent and hydroxyapatite-doped plastics.⁶⁴ These calibration phantoms can either be placed underneath patients (example in Fig. 1) as they are scanned on clinical CT systems to provide an external reference for determining the HU expected for various concentrations of bone^63,64 or a quality control phantom can be scanned at another time to measure BMD using an asynchronous technique.⁶⁵ Examples of phantom-less techniques include using muscle and fat to calibrate the HU of bone^66,67 or using the CT number of trabecular bone directly.^68,69 Using internal calibrations results in errors of up to 1.1% in vertebrae relative to phantom-calibrated scans, while using uncalibrated CT numbers alone is sensitive to variations in tube voltage and body habitus.⁷⁰

FIG. 1.

Example of synchronous scanning of a bone mineral density calibration phantom (white arrow) as it would be positioned under a patient

While DEXA provides a T-score — the number of standard deviations from the average BMD of a healthy 30-yr-old — to determine whether BMD indicates the presence of osteopenia or osteoporosis, spinal CT BMD measurements cannot be directly compared to DEXA-derived areal BMD. However, CT of the proximal femur can be reconstructed to simulate a projection radiograph in the orientation of a DEXA acquisition from which areal BMD can be compared directly in the pelvis,⁷¹ and the two modalities have been shown to correlate very well.⁷² CT-derived BMD of the proximal femur scanned concurrently with a calibration phantom could explain up to 76% of variance in femoral bone strength compared to 69% for DEXA.⁷³ The American College of Radiology (ACR) practice guidelines recommend using the diagnostic criteria in Table II to avoid overestimating the fracture risk of patients from spinal CT⁷⁴ as well as following individual patients on the same CT unit if possible.⁷⁴ It is expected that age related bone density loss in postmenopausal women is between 0.7 and 1.9% per year in white populations,^75,76 but DEXA measurements do not differentiate between losses in cortical versus trabecular bone. Overall, recent studies of quantitative BMD measurements with CT have shown that the coefficient of variation is low when CT operators are well trained, ranging from 0.6 to 3.3%^66,77; in comparison, the coefficient of variation for DEXA measurements is between 0.5 and 2.5%.⁷⁷

Table II.

Spinal quantitative computed tomography bone mineral density (BMD) criteria for degenerative diseases. Table adapted from the American College of Radiology guidelines for performance of musculoskeletal quantitative computed tomography.⁷⁴

Trabecular spine BMD range	World Health Organization diagnostic category
>120 mg/cm3	Normal
80–120 mg/cm3	Osteopenia
<80 mg/cm3	Osteoporosis

Measurements of BMD with DECT may be more accurate than those measured by DEXA and conventional CT, particularly in instances where phantom calibration is asynchronous. The initial DECT BMD studies were published in the 1980s,^78–80 but have recently gained popularity as DECT examinations have become more common clinically. In a conventional CT image of trabecular bone, many voxels contain a mixture of bone marrow (primarily adipose) and trabeculae, which ultimately leads to inaccurate CT numbers that may artificially lead to trabecular bone volume uncertainties of up to 15%.^7,81 In comparison, DECT with three-basis material decomposition comprised of calcium (or hydroxyapatite), adipose tissue, and soft tissue has been shown to minimize the dependence of BMD assessment on the fat percentage of the marrow (R² = 0.05 for DECT vs. R² = 0.56 for SECT, where lower correlation is desirable),⁸² and improve the accuracy and precision of phantom-less imaging techniques.⁸³ Wesarg et al.⁸⁴ demonstrated that DECT-based BMD also shows better correlation with mechanical force measurements relative to DEXA, with R² of 0.82 and 0.48, respectively. While DECT material decomposition is a very promising method for calculating BMD, more work is required to determine the coefficient of variation in a human population. Additionally, hydroxyapatite or calcium concentration measurements from material decomposition algorithms may vary between vendors,^85,86 thus requiring cross-calibration between different DECT platforms.

2.B. Hepatic steatosis (fatty liver disease)

Hepatic steatosis, also known as fatty liver, is the primary symptom of nonalcoholic fatty liver disease (NAFLD), which has an estimated prevalence of approximately 25% worldwide.⁸⁷ common etiologies include, but are not limited to, alcohol consumption,⁸⁸ hepatitis C,^89,90 and use of certain medications⁹¹ — including many chemotherapies^92,93 and corticosteroids.⁹⁴ Nonalcoholic fatty liver disease is also associated with metabolic syndrome, a set of risk factors for cardiovascular disease and type II diabetes that include obesity, hypertension, elevated triglycerides, and insulin resistance.^95,96 Nonalcoholic fatty liver disease is a risk factor for liver cirrhosis^97,98 and hepatocellular carcinoma (HCC).^99,100 In addition to the disease burden caused by NAFLD, hepatic steatosis is also associated with poor outcomes for surgical interventions in the liver.^101–103 The current gold standard for the quantification of liver fat with imaging is magnetic resonance imaging (MRI) Dixon-based sequences, which allow for measurement of the proton density fat fraction (PDFF).^104–107 This has largely replaced MR spectroscopy,^108–110 which is time consuming and typically is performed using a single large voxel that makes it prone to both volume averaging and sampling error. To date, QIBA has not yet published a profile on quantification of hepatic steatosis with CT.

With conventional CT, fatty liver is best diagnosed on non-contrast examinations¹¹¹ because contrast-enhanced CT has limited sensitivity caused by variability in contrast administration, timing, and variations in perfusion. Kodama et al. demonstrated this by showing pathologic fat content had a higher correlation with unenhanced CT HU (R² = 0.649) than contrast-enhanced CT HU (R² = 0.516).¹¹² Other groups have converted CT HU measurements directly to fat fraction using in-scan calibration phantoms, and found that calculated adipose content using the calibration curves were correlated with liver fat upon biopsy (R² = 0.689).¹¹³ In phantoms, Pickhardt et al. also found that CT HU at 120 kVp measurements were highly correlated with MRI-PDFF (R² = 0.828), but single-energy CT tended to underestimate the liver fat content when iron overload was present¹¹⁴ since higher iron content increases the background attenuation in the tissue. Overall, fatty liver is typically diagnosed with CT when the liver attenuation is either below 40 or 10 HU less than the attenuation of the spleen^115–117 (Fig. 2), which does not aim to quantify the fat percentage, but rather indicates the presence or absence of hepatic steatosis.

FIG. 2.

Single-energy computed tomography methodology for diagnosis of hepatic steatosis utilizes regions of interest in the liver and spleen, shown here in a 61-yr-old male seen for restaging of a pancreatic neuroendocrine tumor. The liver attenuation in this case is both below 40 HU and more than 10 HU lower than that of the spleen, indicating the presence of diffuse fat within the liver tissue.

Dual-energy CT with three-material decomposition may also provide additional accuracy for fat segmentation and quantification tasks since it is possible to estimate the fat fraction within the tissue directly (Fig. 3). Several studies have shown that unenhanced DECT using a water-fat material decomposition algorithm alone does not significantly increase the correlation of CT fat fraction and MR-PDFF fat fraction versus single-energy CT.^118–120 However, in a 2016 study by Hyodo et al., a three-material decomposition model based on fat, liver tissue, and contrast agent (iodine) was used in phantoms to accurately assess fat content. They found that while SECT fat estimates demonstrated a dependence on the size of the phantom, DECT fat quantification was largely unaffected by a moderate change in phantom size.¹²¹ The same group assessed the algorithm in patients and found a statistically significant proportional bias in the fat fractions measured on DECT vs. MRS,¹²² which may have been a result of either underestimation of fat volume fraction by DECT, overestimation of fat content by MRS, or variations in iron levels that were associated with the steatosis. These algorithms are also useful for differentiating between lipid-poor and lipid-rich abdominal lesions in the adrenal glands.^123,124 Dual-source DECT has been shown to separate hepatic iron and fat in phantoms^125,126 with a fat, soft tissue, and iron decomposition; however, this algorithm did not provide a percent fat fraction, so a calibration between a virtual non-iron pseudo-HU and pathological fat would be required to directly measure the amount of fat present.¹²⁷ Overall, DECT has shown to have significant advantages over conventional CT for the estimation of fat content in the presence of either iodinated contrast or iron overload in liver tissue.

FIG. 3.

Fatty infiltration of the liver demonstrated on a dual-energy computed tomography fat map in a 75-yr-old woman with pancreatic adenocarcinoma following chemotherapy. The calculated fat fraction for regions of interest in the liver (20.1%) and subcutaneous fat (94.0%) are shown as a percent volume.

2.C. Renal Cell Carcinoma

Renal cell carcinomas (RCCs) represent 80–85% of primary neoplasms in the kidney, and it is estimated that 65 340 people in the United States were newly diagnosed with renal malignancies in 2018, with an additional 14 970 dying of the disease.¹²⁸ There are several imaging criteria that may indicate that a renal lesion is malignant, including the absence of a significant fatty component, contrast enhancement of 15–20 HU,^18,129 cystic components with enhancement (Bosniak types III and IV),^19,20 and tumor complexity. There are a variety of RCC subtypes that represent different tumor morphology, cellular origin, and growth patterns. These subtypes, with their relative incidences, include clear cell (75–85%), papillary or chromophilic (10–15%), chromophobe (5–10%), and oncocytic (3–7%).^130,131 Additionally, fat-poor angiomyolipomas and oncocytomas complicate the differential diagnosis of RCC with imaging alone. While CT imaging is common for renal disease, QIBA has not yet published a profile regarding quantitative analysis of CT number or iodine concentration for differentiation of these lesion types.

Evaluation of mean CT number alone has been shown to lack specificity for differentiating fat-poor angiomyolipoma and oncocytoma from various subtypes of RCC.^132,133 In order to differentiate between these etiologies with conventional CT, quantitative techniques such as texture analysis are currently being investigated. Sasaguri et al. found that CT attenuation and histogram skewness on both the corticomedullary and nephrographic phases were able to differentiate oncocytoma from clear cell RCC (AUC = 0.82) and papillary RCC (AUC = 0.95)¹³⁴ in small lesions; while they found that subjective lesion inhomogeneity was significant, entropy, a texture feature used to quantify pixel randomness, did not reach the level of significance in a multi-variate analysis. Hodgdon et al. evaluated patients with fat-poor angiomyolipoma and RCC and found that models based on various combinations of three unenhanced CT texture features could achieve an AUC between 0.85 and 0.89, with quantitative measures of homogeneity and entropy having the highest predictive power for diagnosis of RCC.¹³⁵ These studies have begun to suggest that quantitative measures of inhomogeneity may be more sensitive than subjective metrics assigned by a radiologist.

DECT has been shown to be particularly valuable for renal imaging, with applications for differentiating subtypes of RCC from cysts^136–138 as well as classifying renal stone composition.¹³⁹ Renal lesions under 4 cm in diameter are a frequent incidental finding on clinical imaging.^134,140 Multiple quantitative metrics, including measured iodine concentration,^{137,141–143} monoenergetic attenuation,^144–146 and effective atomic number (Z_eff)¹⁴⁷ have improved the differentiation of benign and malignant renal lesions. One study by Manoharan et al. is particularly notable because it demonstrated non-inferiority of DECT iodine quantification compared to conventional CT for the evaluation of renal masses in a prospective, randomized controlled trial despite a lower radiation dose with the DECT exams.¹⁴¹ Several studies have demonstrated that iodine quantification was able to distinguish between subtypes of renal cell carcinoma, including clear cell and low-enhancing papillary variants, as well as fat-poor angiomyolipomas.^148–151 Both Chandarana et al. and Mileto et al. demonstrated that 0.5 mg/mL iodine in a renal lesion was sufficient to distinguish non-enhancing from enhancing renal lesions using dual-source DECT^137,138; furthermore, the second group also found an optimal threshold of 0.9 mg/mL for differentiating papillary and clear cell RCCs.¹⁴⁸ Counter to this, Kaza et al. found that the optimal threshold between enhancing and non-enhancing RCCs was 2.0 mg/mL on a fast kVp switching DECT system.¹³⁶ Some of this difference may be attributed to differences in contrast timing between the scan protocols on the two DECT systems; as a result, there is an opportunity for protocol standardization to minimize variance across DECT implementations by different manufacturers.

2.D. Radiotherapy treatment planning

CT numbers have been used extensively since the introduction of the modality to quantify the attenuation of nonuniform tissue for dose calculations in radiotherapy.^152,153 CT numbers are used to estimate the electron density of tissue for photon radiotherapy¹⁵⁴ or proton stopping power for proton therapy.¹⁵⁵ Historically, tissue substitutes were used to calibrate the CT number to specific electron density^154,156 using a bilinear fit, but this technique was dependent on the location of the material of interest within the patient, the accuracy of the CT number itself, and the scanner model.¹⁵⁷ Schneider et al. developed the stoichiometric calibration to more accurately determine the electron density of biological tissues, which improved the mean error in calculated electron density from 2.5% with the tissue substitute method to 0.5% and minimized the number of pixels with more than a 2% deviation from the true electron density.¹⁵⁷ Nakao et al. found that the typical dose errors resulting from day-to-day differences in CT constancy were typically within a range of −0.9 to 0.1% for 6 MV lung plans, and between −1.0 and −1.6% for 10 MV prostate plans¹⁵⁸; this demonstrates the dependence of the electron density conversion on depth originally found by Kilby et al.¹⁵⁹ However, there remains some subjectivity in the application of the stoichiometric method due to the number of line segments the users choose to fit to the calibration data.^160,161 Furthermore, while convolution- or superposition-based treatment planning algorithms only require tissue electron density as an input, Monte Carlo-based treatment planning requires additional material information such as the elemental mass percentages^162–165 to be calculated from the CT data.

Dual-energy CT provides the ability to calculate Z_eff and electron density (q_e), which is advantageous for radiotherapy treatment planning. In particular, the accuracy of the CT-calculated proton stopping power ratio has a direct impact on the proton range estimate, which is critical because there is a sharp dose gradient at the Bragg peak.¹⁶⁶ In addition to the dose gradient, there is also a demonstrated change in the proton LET along the proton range,^167,168 which directly impacts the deposited dose. The stoichiometric method has recently been adapted for DECT data, and it has been shown to improve the accuracy of proton stopping power calculations in validation studies.^169–173 Zhu et al. found that error in stopping power calculated with SECT in a phantom of known composition was as high as 12.8% while DECT resulted in only a 2.2% error, which resulted in dose errors of 7.8% and 1.4% for the SECT and DECT plans, respectively.¹⁶⁶ Dual-energy CT metrics have been evaluated using ground-truth maps by Wohlfahrt et al., who found that several DECT-based stopping power predictions could reliably predict stopping power ratio to within 1% deviation from the median¹⁷⁴ using both electron density/effective atomic number¹⁷⁵ and electron density/photon absorption cross section¹⁷⁶ decompositions. However, they were not able to definitively show that this improved the proton range estimate relative to SECT because the smoothing performed on the SECT and DECT differed relative to the ground-truth map.¹⁷⁴ It is generally agreed that the DECT proton range estimation is lower than that of SECT-based calculations,^170,174,177 but clinical significance has yet to be determined. More recent studies have suggested that DECT may be more accurate and result in lower proton range errors when the beam is in the presence of bone.¹⁷⁸

As discussed previously, it is difficult to accurately determine material characteristics using conventional CT for Monte Carlo treatment planning, both for photon and proton radiotherapy. Several studies have used DECT to estimate material compositions for radiotherapy purposes, including estimations of the water-equivalent path length of protons¹⁷⁹ and tissue oxygen and carbon percentages¹⁸⁰ in addition to calculation of Z_eff.¹⁸¹ An early study by Bazalova et al. reported that Z_eff and electron density estimates from DECT were 3.7% and 3.1%, respectively, when simulating DECT acquisitions at 100/140 kVp with 9 mm of additional aluminum filtration.¹⁸² Using a clinically available second generation dual-source system, Landry et al. found that DECT Z_eff quantification accuracy was within ±10% across a range of tissue substitutes, while electron density accuracy was within 2.5%.¹⁸¹ For low-energy brachytherapy applications, additional optimization of the DECT reconstructions, including the use of iterative reconstruction algorithms can further reduce the standard deviation in Z_eff and electron density and results in maximum dose calculation errors of 6% compared to 21% for SECT.¹⁸³ Dual-energy CT also has been shown to reduce metal artifacts in the presence of implants¹⁸⁴ and metallic brachytherapy seeds,¹⁸⁵ which allows for improved tissue identification¹⁸⁶ and electron density estimation^187,188 for treatment planning.

2.E. Radiomics

Radiomics involves evaluating images at the voxel level to extract quantitative image features (i.e., texture). These radiomics features are then used to build predictive models for overall survival, loco-regional control, or to classify tumors into groupings that may direct treatment. One example of this is using radiomics data from segmented non-small cell lung cancer tumors to stratify patients into varying outcome brackets¹⁸⁹; in this study, an example of a textural feature that was prognostic of survival was tumor compactness, a measure of how spiculated or smooth the tumor may appear. Other models may utilize texture features to differentiate between benign and malignant tumors.

The radiomics features used in most studies are those that come from intensity histograms (e.g., variance, kurtosis, skewness), gray level co-occurrence matrices, defined by Haralick,¹⁹⁰ gray level run length matrices defined by Galloway and Tang in 1975 and 1998, respectively,^191,192 and neighborhood gray tone difference matrices defined by Amadasun and King.¹⁹³ Some additional categories that branch off of these original feature categories have been developed in the past ten years, such as gray level size zone matrix.¹⁹⁴ Recently, the term radiomics has also been used for deep learning approaches that have the same end goal, such as predicting overall survival in cancer patients.

The typical radiomics workflow is comprised of imaging, segmentation, feature extraction, and statistical analysis or cross-validation (Fig. 4). Within this process, there are many confounding factors that could add noise to data sets. Many specific imaging protocol parameters have been investigated to determine their impact on radiomics feature variability, including image thickness, the use of iterative reconstruction versus filtered backprojection, and tube current.^195,196 A recent study combined much of this information and devised a harmonized protocol across different manufacturers and demonstrated that protocol standardization could reduce interscanner variation by over 50% compared to the use of local protocols.¹⁹⁷ However, application of this harmonized protocol has not yet been used for analysis of patient studies. The next radiomics workflow step involves segmentation of the tissue or lesion of interest. This can be performed using CT simulation scans of radiotherapy patients with contours performed by the treating radiation oncologist or application of commercially available tools.¹⁹⁸ Feature extraction from the segmented regions can then be performed using a variety of different software packages, including commercial, open-source, and in-house software.¹⁹⁸ Preprocessing, such as smoothing, is sometimes applied to images before feature extraction. The Image Biomarker Standardization Initiative has attempted to help align features from different software by providing clear definitions and test data sets to benchmark extracted feature values.¹⁹⁹ However, there are still differences in features extracted from different software packages.²⁰⁰ The final step, statistical analysis, can also have potential pitfalls. One study showed that the statistical methodology of some early studies in radiomics could produce spurious correlations.²⁰¹ In each radiomics step, there are areas that still need to be standardized, and these efforts are currently underway.²⁰² Radiomics has shown promise, in particular in stratifying cancer patients into high or low risk for survival.²⁰³ Incorporating these standardization efforts into more studies will help make studies applicable to the larger community and draw radiomics one step closer to clinical applicability.

FIG. 4.

General workflow for development of radiomics-based prediction models. Patients are imaged, and the volume of interest is segmented either manually or automatically. Quantitative radiomics features are extracted from the volume and used along with other clinical or genomics data to develop a statistical model for decision support. Figure reproduced from reference.²⁰⁴

2.F. Potential applications of photon-counting CT

There are several promising clinical applications of quantitative PCD-CT. Because the existing PCD-CT systems are in the preclinical stage, most studies have been performed in phantoms, animals, ex vivo tissue, or cadavers. However, preliminary studies show that PCD-CT is particularly promising for quantitative imaging due to potential improvements in material characterization and increased spatial resolution.

Fat and lean tissue quantification has been evaluated in ex vivo tissue phantoms,²⁰⁵ and phantom and animal studies have shown that calcium concentration can be accurately measured. Accurate calcium quantification combined with assessment of soft tissue composition could be highly relevant to BMD, atherosclerotic plaque²⁰⁶, and hepatic steatosis evaluations. Several studies have shown that hydroxyapatite and calcium oxalate stones can be differentiated with PCD-CT if the energy bins are chosen correctly.^207–209 These studies demonstrate the capabilities of PCD-CT material differentiation, which can be further exploited with development of novel, targeted contrast agents (See Novel Contrast Agent Development).

A special capability of PCD-CT is simultaneous acquisition of high-resolution and multi-energy images, which can’t be achieved at the same time with current commercial CT systems using energy integrating detectors. In quantitative tasks that do not require material decomposition, such as lung density and tumor volumetry, PCD-CT has the ability to provide improvements in spatial resolution that could decrease volume averaging at the boundaries of lesions and improve density measurements.^34,210–214 Additionally, for quantitative tasks that were not feasible with conventional CT due to artifact or high noise levels, PCD-CT allows users to limit the effects of electronic noise, beam hardening artifacts, and metal artifacts through selection of energy thresholds for reconstruction.^{31,38,213,215–218}

3. LIMITATIONS TO CT QUANTITATION

3.A. Conventional CT

There are several limitations to the quantitative use of conventional CT. While the in-plane spatial resolution is high, material differentiation is challenging due to acquisition techniques using a single polychromatic beam. As a result, tissues may present with similar attenuation levels despite having different elemental compositions. For example, a calcified lesion may be indistinguishable from a lesion enhanced with iodinated contrast. This limits the ability of conventional CT to quantify BMD due to the presence of fat in the bone marrow that may lower the overall attenuation, and can limit image contrast when identifying lesion boundaries for volumetric measurements. Other confounding factors such as variation in scanner model²¹⁹ and geometry, scan protocol, patient positioning²²⁰ and size,^221,222 and analysis methods may induce significant bias into measurements, and must be controlled over the course of patient treatment. This is obviously not possible in all cases, so it is beneficial for multicenter trials and meta-analyses to calculate the estimated inter- and intra-patient variability.

In particular, patient positioning contributes to image quality issues which may impede diagnosis as well as quantification with conventional CT. When patients are positioned at the wrong table height, the lack of appropriate energy and flux modulation at the bowtie filter may induce beam hardening artifacts and cause increased image noise. Since spatial resolution degrades away from isocenter, mispositioning may also result in poor visualization of fine structures such as small blood vessels or bony detail.²²³ Poor patient positioning can also lead to issues with tube current modulation due to magnification factors during the localizer scan, causing the dose to either increase or decrease depending on the direction of the positioning error.²²⁴ In an anthropomorphic phantom, Szczykutowicz et al found up to a 15 HU drop in CT number compared to proper positioning in the posterior mid thorax region when the patient was positioned 4 cm above isocenter; absolute errors increased to 20 HU or greater when the phantom was positioned more than 6 cm above isocenter.²²⁰ Relative noise also increases when patients are imaged away from isocenter.^220,225 Large patient size also exacerbates these issues since there is already a larger risk of beam hardening artifact and truncation artifacts in this patient population as well as higher noise. Hardware solutions such as 3D cameras to assist the technologist in positioning the patient are now commercially available, and initial studies demonstrate that their use significantly lowers the deviation between patient isocenter and scan center.^226,227

3.B. Dual-energy CT

There are several existing limitations to quantification tasks with DECT. First, while all vendors provide two-material decomposition, three-material decomposition is only offered by a subset of vendors and for limited material combinations. For example, one vendor offers users choices of materials for two-material decomposition, with reconstructions limited to pairs of water, iodine, calcium, and hydroxyapatite (or cortical bone). In comparison, another vendor offers both two- and three-material decomposition, but the materials are set on an application basis (i.e., liver virtual non-contrast materials are soft tissue, fat, and iodine while a conventional virtual non-contrast is based on a decomposition of soft tissue and iodine). These applications have been useful in both test phantoms and clinical scenarios, but have not yet been established as quantitative biomarkers by the metrology standards set forth by QIBA. Additionally, the problem of material decomposition with three or more basis materials is ill-posed because only two energy spectra are used as inputs to the algorithm. To avoid this problem, volume conservation of materials within a voxel is typically assumed, but this assumption may not hold for materials in solution, requiring other constraints such as mass conservation.²²⁸

Substantial work has been done to improve material decomposition algorithms for DECT over recent years. Liu et al first described the use of mass conservation in three-material decomposition, and the accuracy depended strongly on the difference in the dual-energy ratios of the materials being separated.²²⁸ Several studies have evaluated the use of more than two basis materials in material decomposition, which have the advantage of modeling complex biology or multiple contrast agents within the same object.^120,229,230 Since the accuracy of the CT number impacts material decomposition accuracy as well,²³¹ algorithms have been developed to correct beam hardening for the purposes of DECT reconstruction.²³² Other groups have focused on improving the noise characteristics of material decomposition images with iterative reconstruction^229,233 and machine learning.²³⁴

While DECT has enabled many advances in quantitative CT, there are barriers that remain. The spectral separation is inherently limited in several DECT implementations, including dual-layer detector and split-filter systems. Because these systems both use a single polychromatic beam, the average energies of the two spectra used to compute the material decomposition are relatively similar compared to what is possible on a rapid kVp-switching or dual-source system. The spectral overlap increases the image noise and increases variability between measurements, raising the limit of detection for materials, particularly in large patients. For all DECT systems, the lowest kVp spectrum available is generally limited to 70 or 80 kVp because photon starvation becomes an issue in even average-size adults at energies below this threshold. In brain imaging with DECT, beam hardening in the skull base and temporal lobe can alter the CT numbers before they are entered into the material decomposition algorithm. Furthermore, the assumptions made in the DECT material decomposition process do not enable quantitation of materials that have K-edges within the diagnostic energy range, which limits the clinical utility of many novel contrast agents.

3.C. Photon-counting CT

While PCD-CT is a very promising technology, there are many technical limitations that must be addressed before it can be made readily available in a clinical setting. The first is that developing PCD that can handle the high count rates required for fast CT acquisitions is challenging. Pulse pileup occurs when multiple x rays are incident on a detector element within a very small time window and registered as a single, high-energy photon. As count rates increase, this phenomenon is more likely and skews the detected spectrum. Additionally, count-rate drift due to crystal defects can alter the signal output from individual detector elements, which may lead to ring artifacts.²³⁵ Other nonideal detector response issues can alter the detected spectrum as well, including stochastic generation of electron/hole pairs, incomplete charge collection, charge-sharing, cross-talk between two detector elements caused by x-ray incident near the border of two adjacent detector elements, and fluorescent photons (K-escape) released in an element that are absorbed in adjacent detectors and cause charge cascades.^28,236–238 Each of these examples causes a single photon to be read as two or more separate events in different locations, which blurs the image spatially and lowers the energy resolution of the detector. Thus, the expected detector response becomes distorted and results in imperfect spectral separation between the bins.^28,239 Without correction, the estimated attenuation coefficient of water is underestimated, which causes cupping artifacts²³⁹ in addition to causing biases in the material decomposition algorithms.²⁴⁰ A wide variety of solutions have been proposed for these issues, including detector hardware,^241,242 improved bowtie filters to optimize photon flux,^243,244 and image reconstruction algorithms.^238,245,246

4. FUTURE APPLICATIONS FOR QUANTITATIVE CT

4.A. Novel contrast agent development

Nonionic iodinated agents are the primary form of exogenous contrast used in CT imaging, but there is a small risk of allergic-like reactions ranging from mild (uricaria, nausea, pruritus) to severe (anaphylactic shock) in patients with certain risk factors such as asthma.^247–249 The rate of acute allergic-type reactions to nonionic, low-osmolality iodinated contrast has been estimated at 0.2–0.7%.^250–252 Barium sulfate is also administered orally for certain exams to provide positive contrast in the gastrointestinal tract,^253,254 and xenon-enhanced CT has been used to assess lung ventilation²⁵⁵ and cerebral blood flow.^256,257 The advent of DECT, and particularly PCD-CT, has resulted in increased interest in the use and development of contrast agents using materials that have dual-energy ratios (the HU ratio between low and high kVp) and/or K-edges that differ from iodine. Lambert et al. showed that certain high-Z materials can be differentiated using DECT based on their dual-energy ratios.²⁵⁸ This algorithm was validated in phantoms of coronary arteries with simulated calcific plaques, which were filled with either tungsten or iodine contrast; overall, the error in calcium scores with tungsten contrast agent were 16.3% and 6.4% for DECT material decomposition and their methodology. Respectively, the errors were 39.8% and 9.7% when using iodinated contrast with material decomposition and this method.²⁵⁹

There are several high-Z materials that have K-edges between approximately 40 and 100 keV that are of interest for K-edge imaging with PCD-CT. This technique uses material decomposition with energy bins specifically placed below and above the K-edge of a material for more accurate quantification, as demonstrated in Fig. 5. This can be achieved using filtration on conventional CT systems, as originally conceived by Riederer and Mistretta,²⁶¹ or through optimal choice of energy bins in PCD-CT. Figure 6 shows how the attenuation of gadolinium in phantoms dramatically increases with an energy bin placed just above the K-edge of 50.2 keV. While this is useful from an imaging standpoint, the main drawback of the use of gadolinium as a CT contrast agent is that the concentrations required may leave patients at a higher risk for nephrogenic systemic fibrosis; additionally, several studies have demonstrated gadolinium deposition in deep gray matter of the brain following multiple MRI studies.^262,263 Therefore, significant effort is ongoing for contrast agents based on gold, tungsten, ytterbium, and tantalum (K-edges shown in Table III).²⁶⁴

FIG. 5.

K-edges for iodine (33.2 keV) and tungsten (69.5 keV) allow for K-edge imaging. In energy bin 1 (blue), positioned just below the tungsten K-edge, iodine signal dominates, while in bin 2 (yellow), tungsten signal will be the primary contributor to contrast.

FIG. 6.

Representative (a) grayscale computed tomography image slices of a calibration phantom comprising of 0, 5, 10, 15, and 20 mM gadolinium concentrations showing example VOIs (boxes), and (b) cropped grayscale images of VOIs spanning the full range of gadolinium concentrations, for each energy bin of the photon-counting detector, demonstrating the increased attenuation of gadolinium in energy bins greater than the K-edge (50.2 keV). Note that all grayscale intensities were converted to HU. Figure reproduced from Curtis et al.²⁶⁵

Table III.

K-edge energies for current and experimental computed tomography contrast agents. Data from Deslattes et al.²⁶⁰

Element	K-edge (keV)
Iodine	33.2
Barium	37.4
Cerium	40.4
Gadolinium	50.2
Ytterbium	61.3
Hafnium	65.4
Tantalum	67.4
Tungsten	69.5
Osmium	73.9
Platinum	78.4
Gold	80.7
Bismuth	90.5

Gold nanoparticles have been utilized extensively for a wide range of imaging applications, including both CT and MRI. While it does increase vascular contrast on conventional CT,²⁶⁴ other novel contrast agents provide more contrast in the vasculature and organs at substantially lower cost. However, tumor-specific imaging with targeted^266–268 or theranostic agents²⁶⁹ could be indications that justify the cost.

Tungsten has been used in a dual-contrast dual-energy imaging scheme with iodinated contrast in bowel imaging, and is more promising for this application than bismuth.^270,271 Tungsten has also been demonstrated as a theranostic agent,^272,273 acting as a contrast agent for PCD-CT and an agent to enhance photothermal therapy. However, a potential disadvantage of using tungsten-based contrast agents in CT is that there is low x-ray fluence just above the two K-edges due to self-filtration in the tungsten anode that could potentially reduce image contrast relative to other nanoparticle formulations.

Ytterbium was proposed as a CT contrast agent by Unger and Gutierrez in 1986, and it was found to be visually more dense than iodine at comparable concentrations.²⁷⁴ More recent studies have utilized ytterbium nanoparticles, which are particularly useful for vascular imaging due to their longer circulation time in the blood pool when pegylated.^275,276 Pan et al. have also developed a stable Ytterbium lipid-encapsulated nanocolloid²⁷⁷ that could accurately identify non-targeted ytterbium within a mouse heart with PCD-CT. The long-term goal is to develop a compound targeted to fibrin, which may be advantageous for detecting unstable atherosclerotic plaques in vivo.²⁷⁷ Despite these potential advantages, the amount of ytterbium currently being produced worldwide would not be sufficient for frequent medical use in a large population.²⁶⁴

Tantalum has similar attenuation properties to ytterbium on conventional CT and has been investigated as a CT contrast agent in multiple forms. Like ytterbium, tantalum-based nanoparticles may be useful for vascular imaging, and several studies have demonstrated biocompatability^278,279 and have been validated in porcine models²⁸⁰ with various tantalum oxide (TaCZ) nanoparticle formulations. Other researchers have focused on the possibility of using tantalum oxide cores with various nanoparticle shells to create multi-modality imaging agents compatible with CT, MRI, and either photoacoustic or luminescence²⁸¹ imaging. Tantalum can also be conjugated with doxorubicin as a theranostic agent, which may enable the quantitative tracking of cancer therapies, the feasibility of which has been shown with DECT imaging of platinum-based chemotherapeutic agents. Additional studies have demonstrated a tantalum shell as a simultaneous dual-modality imaging agent, a vessel for chemotherapeutic agents, and a heat sensitizer to treat esophageal malignancies.^282,283 In combination with PCD-CT, tantalum oxide has tremendous potential to quantify drug delivery and improve the delivery of thermal therapies as well as acting as a vascular agent in conventional CT.

4.B. Radiotherapy applications

DECT acquisitions of contrast-enhanced CT simulation exams could provide both contrast-enhanced CT images for optimal physician contouring along with virtual non-contrast CT for dose calculation as a replacement for a true non-contrast scan. However, dose calculations based on virtual non-contrast-enhanced scans still needs to be validated on a large patient cohort with a direct comparison to conventional non-contrast CT. Furthermore, advanced synthesized monoenergetic reconstructions of dual-energy scans have the ability to provide low noise and reduced artifacts and improved image contrast relative to conventional CT by combining the low-frequency data from the low-energy image set and the high-frequency data from the high-energy image set.²⁸⁴ Monoenergetic images have been shown to improve image quality over a blended CT image equivalent to 120 kVp.^285,286 Both Kaup²⁸⁶ and Wichmann²⁸⁵ tested 40, 60, 80, and 100 keV synthesized monoenergetic images for viewing lesions in the lung and head and neck, respectively. Both studies concluded that lesion contrast was greatest at 40 keV, but the contrast-to-noise ratio was optimized at 60 keV. Optimal scan parameters for image quality and dose calculations vary across studies, and further work is needed to determine best practices. Additional studies are required to objectively quantify the effect this improved image quality has on physician contours of tumor and normal tissue structures across a large patient population.

Studies have shown the benefit of DECT^174,287 and photon-counting CT²⁸⁸ for improved calculation of stopping power ratio in proton therapy planning. However, the clinical implementation of DECT for proton therapy is widely varied across institutions, making its true clinical impact difficult to discern.²⁸⁹ Future work is needed to optimize and standardize the use of DECT for stopping power ratio calculations in proton therapy applications. A step-wise approach to the implementation of DECT for proton therapy planning has been proposed, which would begin with transitioning from conventional CT to virtual monoenergetic images from DECT, and finally to DECT-based electron density estimates,²⁹⁰ to enable adoption by clinics, commercial treatment planning software, and regulatory bodies.

The ability of DECT to quantify the presence of high-Z materials will make it an important tool for future nanoparticle-enhanced radiation therapy. The radiation dose-enhancing effects of gold nanoparticles has been thoroughly investigated in simulations, in vitro and in vivo small animal studies.²⁹¹ Dual-energy CT and PCD-CT have shown the ability to quantify the distribution of gold nanoparticles in simulations, in vitro and in vivo.^292–294 However, the translation of this technology for human use has been stalled by difficulties targeting the nanoparticles to the tumor²⁹⁵ since many gold nanoparticles rely solely on the enhanced permeation and retention effect of tumors to increase uptake relative to healthy tissues. As an alternative to gold, hafnium-based nanoparticles are currently being investigated in clinical trials.²⁹⁶ Although less work has been done to investigate the quantification of these particles using DECT, there is some evidence that it is possible,²⁹⁷ and hafnium, with a K-edge of 65.4 keV, may be a candidate for K-edge imaging with PCD-CT. Should these particles make it to clinical applications, DECT and PCD-CT could play an important role in quantifying the distribution in the tumor for advanced radiation treatment planning.

4.C. Image-guided intervention applications

There are many opportunities for development of quantitative CT applications for image-guided therapy. Transcatheter arterial chemoembolization (TACE) is a treatment modality primarily for liver tumors in which an interventional radiologist injects chemotherapeutic and embolic agents directly into a tumor’s blood supply under image guidance. In order to visualize the treatment delivery, the chemotherapeutic agent is emulsified with a small amount of lipiodol (iodinated poppy seed oil) prior to injection,^298–300 and the embolization is considered successful if there is complete lipiodol uptake in the tumor without arterial enhancement in the tumor on CT acquired one month post-treatment. One major disadvantage of using lipiodol for TACE monitoring is that the high attenuation of the lipiodol makes it difficult to assess the residual vascularity of tumors with angiography immediately following treatment. In-room DECT or PCD-CT would allow for acquisition of CT angiography using alternative contrast agents that could be separated from the lipiodol using material decomposition algorithms and application of new theranostic agents.

Separation of the TACE drug from the lipiodol following injection is one concern for this paradigm. Tagging the chemotherapeutic agent directly with a highly attenuating material such as bismuth, tungsten, or ytterbium could allow for direct tracking of the drug, and could also be applied to drug-eluting particle therapies. In a recent article by Sofue et al. multi-energy CT was used to quantify the concentration of cisplatin, a platinum-based chemotherapeutic agent, in the presence of iodine using phantoms and a three-material decomposition of cisplatin, iodine, and agar.³⁰¹ While this has not yet been tested in patients, the study demonstrates that drugs containing high-Z materials with dual-energy ratios significantly different from iodine can be identified on imaging using DECT. This effect could be further exploited using K-edge imaging with PCD-CT by tagging drugs with nanoparticle contrast agents. This will enable visualization of the actual patient-specific biodistribution within the tumor, quantification of drug delivery to the tumor and evaluation of new drug-eluting medical devices for pharmacokinetics.

5. DISCUSSION

The use of imaging in medicine has largely been limited to qualitative assessments of anatomy and disease processes. Quantitative CT can play a large part in reducing the subjectivity of imaging interpretation moving forward. Examples of this include bone mineral density assessment, which is increasingly performed on an opportunistic basis with CT, and the use of CT number thresholds for diagnosis, as seen in conventional imaging for hepatic steatosis. Even RECIST criteria can suffer from physician bias in the determination of tumor diameters, and fatty infiltration of the liver may not be uniform and so ROI-based assessments may not characterize the whole liver. A current gap in knowledge is the impact of specific acquisition and reconstruction parameters on quantitative imaging results. Therefore, in addition to the development of new quantitative metrics, there must be steps taken to automate and standardize the way measurements are gathered for particular disease profiles. While this is beginning to be addressed in the field of radiomics,¹⁹⁷ it typically has not been assessed for DECT-based quantification methods. This must be performed on both a small scale (e.g., parameters at a given institution) and large scale (e.g., comparisons between vendors/platforms and multi-institutional studies). Broader standardization of imaging protocols across institutions can help to minimize variability in interpretation and quantitative results, and can be used to better inform clinicians about relevant biological processes. However, there remain barriers to the inclusion of such quantitative information, including the need for structured reporting of data and development of workflows for quantitative analysis.

Photon-counting detector CT is the most recent CT hardware technology to be developed, and human prototypes are now available at select academic centers around the world. The energy resolution of the detector enables better identification and quantification of novel materials, and spatial resolution is greatly improved over conventional CT. These advancements can assist with the characterization of tumors, atherosclerotic plaques, and abnormal vasculature. As this is largely a new topic in the field of medical imaging, there are many opportunities for advancement and application development to prepare the technology for clinical use. Photon-counting detector CT spectral distortions from nonideal detector performance currently minimize the potential advantages of these systems over DECT spectral acquisition, but developing corrections for the spectrum is an active area of research. There are multiple detector technologies available, and comparisons between them will need to be performed as systems are released commercially. Additional reconstruction parameters and image types must be optimized for specific imaging protocols, including the number and spacing of energy bins based on both the contrast agents utilized and the noise profile within each bin that is tolerable for visualization, material decomposition, and quantification. Applications currently may require development in large animal models, cadaver studies, or phantom experiments in order to translate them from the bench to the bedside. The impact of larger image matrices to accommodate the system resolution requires evaluation, and as contrast agents are developed, each will need to undergo clinical trials for safety and efficacy. As with the adoption of DECT, there must also be a significant effort to educate physicians about the differences in expected image characteristics between conventional and PCD-CT, such as reduced image noise and increased spatial resolution.

While development of novel tracers is common for nuclear medicine modalities, there has been very little focus on the use of new CT contrast agents until the development of PCD-CT. Clinical iodinated contrast agents are unable to be targeted in their current form, and therefore do not provide the diagnostic specificity that targeted tracers may be capable of achieving. Unique CT tracers would be particularly valuable for oncologic imaging, as they could enable enhanced detection of both primary tumors and distant metastases, with the potential for improving detection of smaller metastases than would be feasible with conventional CT. Subsequently, there could be improved delineation of gross target volumes in radiotherapy to include additional microscopic disease. In oncology, patients typically receive follow-up imaging every 3–6 months, and DECT has already had a positive impact on monitoring treatment response for many patients during their care, but innovative contrast agents could potentially provide additional and specific diagnostic information to clinicians regarding the characteristics of tumor heterogeneity, the tumor microenvironment, or associated immune and inflammatory responses. The nanoparticle contrast agents described in Section 5.A can all be readily adapted to this new imaging paradigm. However, the majority of novel contrast agents are being developed for use with PCD-CT and have K-edges between 40 and 100 keV, which is disallowed by the assumptions of currently available DECT material decomposition algorithms. In addition to limited applicability to current scanner technology, targeted contrast agents may only be present within diseased tissue in very small quantities. Therefore, high PCD-CT sensitivity is paramount in order to detect these agents and may depend on parameters such as the number of energy bins and the binning of pixels within the detector, and will be an ongoing area of interest as more human PCD-CT prototypes are made available.

6. CONCLUSIONS

The advancements in multi-energy CT demonstrate exciting opportunities for quantitative imaging. This is imperative, as there is a critical need to improve characterization and response assessment of disease and subsets of diseases. Realization of additional enhancements are conceivable with the development of new, targeted contrast agents along with high-resolution material-specific imaging (e.g. PCD-CT K-edge) to further improve specificity and potential new applications, especially translation to image-guided therapy and interventions. Current challenges in quantitative imaging, such as limited assessment of protocol-induced variability in metrics, provide unique opportunities for medical physicists to become involved in translational research, provide physician education, and impact patient care. In summary, the technical advancements of quantitative multi-energy CT provides numerous opportunities for advancement related to biology, drug development and medical subspecialties to help reduce variability in interpretation, and provide valuable insight to physicians and patients.