Prognostication of overall survival in patients with brain metastases using diffusion tensor imaging and dynamic susceptibility contrast-enhanced MRI

Abstract

Objectives:

To investigate the prognostic utility of DTI and DSC-PWI perfusion-derived parameters in brain metastases patients.

Methods:

Retrospective analyses of DTI-derived parameters (MD, FA, CL, CP, and CS) and DSC-perfusion PWI-derived rCBV_max from 101 patients diagnosed with brain metastases prior to treatment were performed. Using semi-automated segmentation, DTI metrics and rCBV_max were quantified from enhancing areas of the dominant metastatic lesion. For each metric, patients were classified as short- and long-term survivors based on analysis of the best coefficient for each parameter and percentile to separate the groups. Kaplan-Meier analysis was used to compare mOS between these groups. Multivariate survival analysis was subsequently conducted. A correlative histopathologic analysis was performed in a subcohort (n = 10) with DTI metrics and rCBV_max on opposite ends of the spectrum.

Results:

Significant differences in mOS were observed for MD_min (p < 0.05), FA (p < 0.01), CL (p < 0.05), and CP (p < 0.01) and trend toward significance for rCBV_max (p = 0.07) between the two risk groups, in the univariate analysis. On multivariate analysis, the best predictive survival model was comprised of MD_min (p = 0.05), rCBV_max (p < 0.05), RPA (p < 0.0001), and number of lesions (p = 0.07). On histopathology, metastatic tumors showed significant differences in the amount of stroma depending on the combination of DTI metrics and rCBVmax values. Patients with high stromal content demonstrated poorer mOS.

Conclusion:

Pretreatment DTI-derived parameters, notably MD_min and rCBVmax, are promising imaging markers for prognostication of OS in patients with brain metastases. Stromal cellularity may be a contributing factor to these differences.

Advances in knowledge:

The correlation of DTI-derived metrics and perfusion MRI with patient outcomes has not been investigated in patients with treatment naïve brain metastasis. DTI and DSC-PWI can aid in therapeutic decision-making by providing additional clinical guidance.

Introduction

Brain metastases are the most frequent central nervous system tumors in adults, with increasing incidence due to improved access to diagnostic imaging methods and prolonged survival from primary neoplasms. ^1–3 It is therefore essential to accurately prognosticate these patients for appropriate therapeutic decision-making. Several scoring systems have been proposed to predict the prognosis of brain metastases. The recursive portioning analysis (RPA) assesses the Karnofsky performance status (KPS), age, controlled primary tumor, and extracranial metastasis; the graded prognostic assessment (GPA) evaluates KPS, age, number of brain metastasis, and extracranial metastasis; and the diagnosis-specific graded prognostic assessment (DS-GPA) considers the same characteristic as GPA in addition to primary tumor histology. ^4–6

It has been reported that the higher the number of quantitative measurements assessed, the less subjective the scoring system is. ^7–9 GPA includes the number of brain metastases, and RPA does not, and GPA is found to be more clinically useful, providing better treatment guidance. ¹⁰ Moreover, RPA was inefficient in patients with more than four metastases ¹¹ and large variability in prognosis was observed within the intermediate RPA II class, ¹² while a lack of power was indicated to be a limitation of the DS-GPA in predicting survival between histologic subtypes. ⁶ Growing evidence suggests that when the prognosis is unclear, clinicians require additional guidance to select appropriate treatment recommendations. Multiparametric MRI including diffusion tensor imaging (DTI) and dynamic susceptibility contrast-perfusion-weighted imaging (DSC-PWI) provides metabolic and physiologic information complementing the morphologic information about metastasis number and location, thereby providing added value to be used as objective biomarkers for prognostication purposes. ¹³

A recent study suggested that combination of apparent diffusion coefficient (ADC) with clinical prognostic markers improves the prognostication of patients with resected brain metastases. ¹⁴ However, ADC, which has the same physiologic significance as mean diffusivity (MD), provides only the information about the magnitude of water molecular motion. DTI is commonly used for brain imaging and provides additional information about the tensor orientation and shape, including fractional anisotropy (FA), linear anisotropy (CL), and planar anisotropy (CP). ^15,16 Previous studies, including from our group, have demonstrated the utility of DTI metrics in differentiating solitary brain metastasis from glioblastomas, ^17–20 as well as for prediction of overall survival in glioblastoma patients. ²¹ These studies indicate that FA, CL, and CP values provide additional information about tumor diffusion characteristics, which may also be helpful to noninvasively assess the microstructure of brain metastasis, strengthening the utility of a single metric ‘ADC’ to better separate risk groups.

Besides DTI, DSC-PWI derived cerebral blood volume (CBV) is a relevant and potent biomarker to assess tumor angiogenesis and microvasculature, ^22,23 and has been demonstrated to be a valid biomarker to predict survival in patients with newly diagnosed glioblastoma – patients with increased relative CBV were associated with poorer outcomes. ^24–26 On the contrary, in a previous histopathologic analysis, patients with brain metastases with high microvascular density and a neoangiogenic vascularization pattern had favorable survival times. ²⁷ However, pretreatment perfusion characteristics of brain metastases are yet to show significant prediction of overall response. ^13,28,29

The combination of both DTI and DSC-PWI derived metrics provides additional information on the intratumoral heterogeneity and has been shown to improve prognostication in high-grade gliomas. ^30–33 Therefore, in the present study, we aimed to investigate the prognostic utility of DTI and DSC- PWI derived parameters in a relatively large cohort of patients with treatment naïve brain metastases. In order to validate the imaging findings, correlative histopathologic analysis measuring stromal component (desmoplastic reaction) was performed in a subset of patients.

Methods and materials

Patient population

This retrospective study was approved by the institutional review board (IRB) and was compliant with the Health Insurance Portability and Accountability Act. The inclusion criteria included a histologically confirmed diagnosis of brain metastasis and availability of conventional and advanced (DTI and DSC-PWI) MRI scans prior to any treatment for brain metastasis.

Based upon the inclusion criteria, 101 patients (mean age = 59, SD±12 years, 49 males/ 52 females) with solitary and multiple brain metastases (two, three, or more than three) were included. Primary cancer was lung (n = 56), breast (n = 15), and other (n = 30). Six patients were excluded from the analysis of DSC- PWI due to the presence of susceptibility artifacts. Clinical data included tumor type, KPS, GPA, RPA, and overall survival, which was measured as the time interval from initial MRI diagnosis of brain metastasis until the date of death (n = 82) or last known clinical encounter, if the patient was alive (n = 19). The details of each patient’s demographic, clinical, and diffusion and perfusion MRI-derived parameters are described in Supplementary Table 1.

MR imaging data acquisition

Diffusion Tensor imaging

Axial DTI data were acquired using 30 noncollinear/noncoplanar directions with a single-shot spin-echo, echo-planar read-out sequence with parallel imaging using generalized autocalibrating partially parallel acquisition (GRAPPA) and acceleration factor of 2. The sequence parameters were as follows: repetition time (TR) / echo time (TE) = 5,000/86 ms, number of excitations (NEX) = 3, field of view (FOV) = 22 x 22cm², matrix size = 128 x 128, in-plane resolution = 1.72 x 1.72 mm²; slice thickness = 3 mm; b = 0, 1000 s/mm²; number of slices = 40; acquisition time 8 min.

Dynamic susceptibility Contrast-Perfusion-weighted imaging

For axial DSC- PWI, a bolus of gadobenate dimeglumine (MultiHance; Bracco Diagnostics, Princeton, New Jersey) was injected with a preloading dose of 0.07 mmol/kg to reduce the effect of contrast agent leakage on CBV measurements. A T2*-weighted gradient-echo EPI was used during the second 0.07 mmol /kg bolus of contrast agent for the DSC- PWI. The injection rate was 5 ml s⁻¹ for all patients and was immediately followed by a flush of saline (total of 20 ml at the same rate). The sequence parameters were as follows: TR/TE = 2000/45 ms; FOV = 22 x 22 cm²; matrix size = 128 x 128; in-plane resolution = 1.72 x 1.72 mm²; slice thickness = 3 mm; BW = 1346 Hz/pixel; flip angle = 90°; EPI factor = 128; echo spacing = 0.83; acquisition time 3 min and 10 sec. Forty-five sequential measurements were acquired for each section. After the preloading dose of the contrast agent, routine sequences were also obtained, including axial T ₁-weighted 3D MPRAGE postcontrast (TR/TE/TI 1760/3.1/950 ms; 192 × 256 matrix size; 1-mm-section thickness; acquisition time 3 min and 10 s) and axial FLAIR (TR/TE/TI 9420/141/2500 ms; 3-mm-section thickness; acquisition time 3 min and 10 s).

Image processing and data analysis

The motion and eddy current correction modules were applied to raw DTI data using in-house developed software (IDL; ITT Visual Information Solutions, Boulder, Colorado). Pixel-wise MD, FA, CL, CP, and spherical coefficient (CS) maps were computed by using the methods described earlier. ^34,35 Leakage-corrected CBV maps were generated by performing gamma-variate curve fitting from DSC-PWI data using NordicICE software (NordicNeuroLab, Bergen, Norway).

The DTI derived maps, CBV maps, and T2-FLAIR images were resliced and co-registered to contrast-enhanced T ₁-weighted images. A semiautomatic approach was used to segment the contrast-enhancing regions of the dominant metastatic lesion by using a signal intensity-based thresholding method as defined previously. ^34,35

Tissue-Based analysis

Representative histologic sections stained with hematoxylin-eosin and Masson’s Trichrome were used to assess the stromal component (desmoplastic reaction) within the tumor by a board-certified neuropathologist (A.N.V.). The staining procedure for Masson’s Trichrome included: slides were deparaffinized and hydrated, thereafter placed in Bouin’s fixative for 1 h in 60 ˚C oven, washed in running tap water for 5 min, rinsed in deionized water, and stained in Weigert’s Iron Hematoxylin for 10 min. Subsequently, slides were washed in running tap water for 10 min, rinsed in deionized water, stained in Beibrich Scarlet- Acid Fuchsin solution for 15 min, and rinsed in deionized water, stained in Phosphotungstic acid for 2 min. Finally, the slides were stained in Aniline Blue solution for 1 to 5 min, rinsed in 1% acetic acid, dehydrated and coverslipped. The amount of tumor stroma was qualitatively assessed as low (scant to little), moderate, and high (stroma comprising a noticeable proportion of the overall tumor volume).

Final data and statistical analysis

DTI metrics and DSC-PWI derived CBV values were quantified from enhancing areas of dominant metastasis segmented semi-automatically, as described above. We used continuous values and various breakdowns in terms of coefficients (Q10, Q25, Q50, Q75, and Q90) for each parameter and percentiles (25th, 50th, and 75th percentile) to separate the patients into two risk groups (short and long-term survivors) in a univariate analysis. The best combination for each parameter from preliminary results was as follows, MD: the lower 10th percentile MD values (Q10) reported as minimum MD (MD_min) and subsequently separated the risk groups above and below the median (50th percentile) of the MD_min. FA/CL/CP/CS: the lower 25th percentile FA/CL/CP/CS values (Q25) and subsequently separated the risk groups above and below the first quartile (25th percentile) of the FA/CL/CP/CS Q25. rCBV: the top 90th percentile rCBV values (Q90) reported as rCBV_max and subsequently separated the risk groups above and below the third quartile (75th percentile) of the rCBV_max. Kaplan-Meier survival curve plots and log-rank tests were used to compare the survival rates using the aforementioned imaging parameters (MD_min, FA Q25, CL Q25, CP Q25, CS Q25, and rCBV_max), as well as clinical variables (number of brain metastasis and site of primary tumor), and clinical scoring systems (GPA and RPA classes) as stratification factors. Subsequently, clinical and imaging variables that demonstrated significant predictive values and trends toward significance from univariate survival analyses were incorporated into multivariate survival analysis using the Cox regression hazard model with backward conditional method. A probabilistic (p) value of less than 0.05 was considered significant. All statistical analyses were performed using a statistical package, SPSS for Windows (v. 18.0; Chicago, IL).

Additionally, a correlative histopathologic analysis was performed in a subset of patients on the opposite ends of the spectrum of DTI metrics and DSC-PWI derived rCBV_max. The DTI metrics were used as a surrogate marker of tumor cellularity and microarchitectural organization ¹³ and rCBV_max for tumor angiogenesis and neovascularization. ²⁷  Group 1 (n = 6; Supplementary Table 1: patients 12, 31, 66, 84, 87, and 94) was selected to represent the best combination of the lowest values of MD_min (impaired mobility of water; high cellularity) and the highest values of FA Q25 and CP Q25 (increased anisotropy; high organization of the tumor microarchitecture), and the lowest values of rCBV_max (reduced angiogenesis; poor vascularity). Group 2 (n = 4, Supplementary Table 1: patients 13, 34, 77, and 81) was selected to have the best combination of the highest values of MD_min (facilitated mobility of water; low cellularity) and the lowest values of FA Q25 and CP Q25 (decreased anisotropy; poor organization of the tumor microarchitecture), and the highest values of rCBV_max (elevated angiogenesis; increased vascularity). Of note, similar to the main cohort, this subcohort of patients also had metastatic tumors from a variety of primary sites (six lung adenocarcinomas, one breast adenocarcinoma, one colorectal adenocarcinoma, one high-grade neuroendocrine carcinoma of lung origin, and one sarcoma of lung origin).

Results

Survival analysis

The Median KPS score was 80, and the median overall survival (mOS) for all patients was 301 days. In univariate analysis, survival differences by GPA class was significant (log-rank p < 0.001. Class 1: 339.35 ± 128.79 days; Class 2: 682.64 ± 100.49 days; Class 3: 1519.86 days; and Class 4: 1553.67 ± 425.63 days) and also for RPA class (log-rank p < 0.001. Class 1: 1533.13 ± 238.61 days; Class 2: 505.25 ± 72.71 days; Class 3: 123.75 ± 30.09 days), indicating a valid cohort. The mOS using tumor primary site (lung, breast, melanoma, kidney, colon, or other) was not significantly different (log-rank p > 0.05), and the number of brain metastasis showed a trend toward significance (log-rank p = 0.09), as independent predictors in univariate analysis.

Determination of long and short-term survivors using MRI data

Kaplan-Meier analyses revealed significant differences in mOS for MD_min (log-rank p < 0.05), FA Q25 (log-rank p < 0.01), CL Q25 (log-rank p < 0.05), and CP Q25 (log-rank p < 0.01), and a trend toward significance for rCBV_max (log-rank p = 0.07). However, CS Q25 (log-rank p > 0.05) did not predict survival (Figures 1 and 2 and Table 1).

Figure 1.

Kaplan–Meier curves showing significantly higher overall survival (log-rank p = 0.013) in patients with MD_min above 50^th percentile, significantly higher overall survival (log-rank p = 0.009) in patients with FA Q25 below 25^th percentile, and significantly higher overall survival (log-rank p = 0.009) in patients with CP Q25 below 25^th percentile.

Figure 2.

Kaplan–Meier curves showing a non-significant difference in overall survival in for CS Q25 (log-rank p = 0.384) and a trend toward significance for rCBV_max (log-rank p = 0.073).

Table 1.

Determination of long and short-term survivors using MRI data

Coefficients	Percentiles	mOS	log-rank p
MDmin	Below the 50th percentile	470.81 ± 82.44	<0.05
	Above the 50th percentile	983.44 ± 153.19
FA Q25	Below the 25th percentile	1269.92 ± 249.96	<0.01
	Above the 25th percentile	565.073 ± 78.75
CL Q25	Below the 25th percentile	1162.31 ± 240.35	<0.05
	Above 25th percentile	622.832 ± 92.68
CP Q25	Below the 25th percentile	1253.06 ± 252.02	<0.01
	Above 25th percentile	571.86 ± 79.64
CS Q25	Below the 25th percentile	584.62 ± 140.18	>0.05
	Above 25th percentile	785.69 ± 111.94
rCBVmax	Below the 75th percentile	853.19 ± 98.71	0.07
	Above the 75th percentile	1126.57 ± 268.98

CL, linear coefficient; CP, planar coefficient; CS, spherical coefficient; FA, fractional anisotropy; MDmin, minimum mean diffusivity; mOS, median overall survival; rCBVmax, maximum relative cerebral blood volume.

Determination of the best-predictive survival model

When statistically significant imaging and clinical independent prognostication parameters (MD_min, FA Q25, CL Q25, CP Q25, RPA, and GPA) and parameters with trends toward significance, such as rCBV_max (log-rank p = 0.07) and number of metastatic brain lesions (log-rank p = 0.09), from univariate analyses were incorporated into Cox proportion hazard regression test, the best-predictive survival model was obtained, which consisted of a combination of MD_min (p = 0.05), rCBV_max (p < 0.05), RPA (p < 0.0001), and number of brain metastasis (p = 0.07) in determining the OS.

Tissue-based analysis

Metastatic tumors showed an overall higher interstitial connective tissue (stromal component) in Group 1 (n = 6, mOS 83.33 ± 12.41 days) as compared to Group 2 (n = 4, mOS 1858.00 ± 707.81 days). The lowest values of MD_min and rCBV_max, along with the highest values of FA Q25 and CP Q25 (Group 1) significantly correlated with elevated stromal content and worse mOS (Figure 3). On the other hand, the highest values of MD_min and rCBV_max, along with the lowest values of FA Q25 and CP Q25 (Group 2) correlated with lower stromal content and better mOS (Figure 4).

Figure 3.

55-year-old female with metastatic adenocarcinoma of colorectal origin, prior to any treatment for brain metastasis, presenting with mOS of 62 days. (A) Contrast-enhanced T ₁-weighted image shows a heterogeneous enhancing lesion in the left inferior frontal lobe. (B) FLAIR images demonstrate surrounding vasogenic edema. (C) Decreased ADC (ADC_min = 0,000692) and (D) very high FA (FA Q25 = 0,12) values from the enhancing part. (E) DSC- PWI with low rCBV from the enhancing region of the tumor (rCBV_max = 6.28; white arrows in E). (F) and (G) Pathology from surgical resection demonstrates high stromal component (blue staining, black arrows in G). H&E stain (F) and Masson’s trichrome stain (G). Histological photos were taken at 100x magnification.

Figure 4.

45-year-old female with metastatic adenocarcinoma of breast, prior to any treatment for brain metastasis, presenting with mOS of 2859 days. (A) Contrast-enhanced T ₁-weighted image shows a ring enhancing lesion delimitating a central area of necrosis in the left temporal lobe. (B) FLAIR images demonstrate surrounding vasogenic edema. (C) Increased ADC (ADC_min = 0,000905) and (D) very low FA (FA Q25 = 0,057) values from the enhancing part. (E) DSC- PWI with elevated rCBV from the enhancing region of the tumor (rCBV_max = 8.64, white arrows in E). (F) and (G) Pathology from surgical resection demonstrates a smaller stromal component (blue staining, white arrows in G). H&E stain (F) and Masson’s trichrome stain (G). Histological photos were taken at 100x magnification.

A quantitative measurement of the stromal component was not performed due to large variability across the slides. Thus, a qualitative assessment of low, moderate and high component was used instead. Assessment of the stromal component was facilitated by Masson’s Trichrome stained sections (blue staining). These findings were consistent within the two groups regardless of tumor site of primary origin.

Discussion

The prognostic value of ADC is well-known for extracranial cancers in predicting survival, response to therapy, and even propensity to form brain metastasis. ^36,37 Primary CNS tumors with low ADC values have also been associated with worse outcomes. ^38–40 Recently, ADC incorporated with standard clinical parameters has shown to improve the prediction of overall survival in patients with surgically resected brain metastasis. ¹⁴ In this study, the MD_min, FA Q25, CL Q25, CP 25, and CBV_max as independent measures separated risk groups in short and long-term survivors, and in a multivariate analysis, the combination of MD_min, rCBV_max, RPA, and the number of brain metastasis was the best predictive survival model, reinforcing the importance of MD_min and rCBV_max as imaging parameters to predict OS and a possible confounding effect of RPA and the number of metastatic lesions. In addition, we demonstrated a correlation between the amount of desmoplastic reaction and mOS, substantiating the DTI metrics and rCBV_max values.

The correlation of DTI-derived parameters and perfusion with patient outcomes has not been investigated in treatment naïve brain metastasis. However, a few studies reported the importance of ADC values in predicting survival in preoperative brain metastasis. ^41,42 These studies reported aggressive tumor behavior in patients with low ADC values. In accordance with these studies, we also observed lower mOS in patients with MD_min below the median (similar physiological significance of ADC), and the MD_min remained statistically significant in multivariate analysis, supporting its prognostication value. However, other studies, ^17–19 including from our group, ²⁰ have demonstrated that the application of more gradient directions (DTI) can provide further tissue characterization and potentially be utilized for prognostication. We, therefore, evaluated several DTI parameters to noninvasively assess the tumor microarchitecture, demonstrating poor mOS in patients with low MD_min (impaired mobility of water) and high FA Q25, CL Q25, and CP 25 values (increased anisotropy). In other words, patients with high cellularity and organization of the tumor microarchitecture comprised short-term survivors. On the other hand, patients with high MD_min (facilitated mobility of water) and low FA Q25, CL Q25, and CP 25 values (decreased anisotropy), that is, low cellularity and disorganization of the tumor microarchitecture, were represented by long-term survivors.

Histopathologic studies have identified dense stromal matrix in primary tumors characterized by increased collagen fibril stiffness and anisotropy, described as a desmoplastic reaction, corroborating tumor growth. ^43,44 In preoperative brain metastasis, Berghoff et al 2013 ⁴² reported a significant correlation of prominent interstitial fibrosis with the semiquantitative DWI signal intensity, resembling the impaired mobility of water molecules in the intercellular space. By extending the quantitative assessment of multiple DTI parameters, we demonstrated higher stromal component in patients with overall worse survival prognosis, represented by the lowest MD_min values and the highest values of FA Q25 and CP Q25. In contrast, patients with the highest MD_min values and the lowest FA Q25 and CP Q25 values presented with low cellular stromal component and had better long-term survival. These results from analyzed DTI metrics may reflect the tumor microarchitecture density and organization, reinforcing the importance of the interstitial space in the pathobiology of brain metastasis and the potential of quantitative DTI metrics to indirectly evaluate histopathological features.

The efficacy of chemoradiation therapy relies on the effective delivery of therapeutic agents and oxygen to the tumor cells. However, delivery of the drug and oxygen is often impeded by abnormal blood vessels and the presence of tumor hypoxia. ⁴⁵ Using MRI-based perfusion parameters such as vascular transfer constant (K^trans, a measure of tumor blood flow and vascular permeability), several previous studies ^46–49 have reported that patients with extracranial cancers exhibiting elevated pretreatment K^trans harbor prolonged OS. In the present study, rCBV_max was found to be a significant predictor in the determination of OS from Cox regression analysis, and brain metastasis patients with higher pretreatment rCBV_max had longer mOS. Our results and those of earlier published reports support the notion that tumors with relatively higher blood flow/volume are associated with increased oxygenation levels resulting in better access to chemotherapeutic drugs and radiosensitivity and thereby associated with improved survival outcomes.

The dense stromal matrix has also been correlated with pancreatic ductal adenocarcinoma with deficient vasculature limiting the delivery of chemotherapy and resulting in resistance to systemic therapies. ⁵⁰ Therapeutic targeting of stromal cells increases intratumoral perfusion and enhances therapeutic delivery. In the current study, the group of patients with the lowest values of rCBV_max (reduced tumor angiogenesis) showed higher stromal component and dismal survival rates. According to a prior histopathological study, ²⁷ small peritumoral edema around brain metastasis was correlated with low microvascular density and neoangiogenic vascularization and was more likely to show brain-invasive growth than tumors with large edema. Therefore, rCBV as a measure of angiogenic activity may indirectly reflect tumor infiltration and predict OS.

There is a growing body of evidence supporting that advanced MRI techniques could potentially be integrated into the widely validated prognostic models for brain metastases patients, ¹⁴ such as RPA or GPA scores, which can be improved with quantitative imaging metrics assessing the metabolic and physiologic information of the tumor. Our results suggest that the incorporation of rCBV_max and MD_min could further enhance these prognostic markers, potentially contributing to therapeutic decision-making. For example, in future, brain metastases patients with a good prognosis can be offered more aggressive treatments, whereas, in patients with poor prognosis, optimal treatment focused on quality of life would be deemed more appropriate.

Despite promising findings, our study was also associated with certain limitations, including the retrospective study design. In addition, histological correlation was only performed on a limited sample size. Furthermore, our data is from a single institution and could benefit from validation in a multiinstitutional cohort, preferably in a larger prospective study.

Conclusion

In conclusion, we reported the prognostic value of DTI metrics and rCBV_max in segregating long and short-term survivors with brain metastasis and their correlation with tissue-based stromal features. Additionally, we reinforced the importance of MD_min and rCBV_max in a multivariate analysis. Further validation in larger prospective studies in a multiinstitutional setting may substantiate the incorporation of advanced imaging techniques into established risk stratification models.