Although diffusion-weighted MRI imaging (DWI) is sometimes considered separately from conventional ceMR, DWI is always acquired as part of the ceMR for brain tumor imaging in clinical practice (40). DWI uses measurements of random water motion to characterize the microstructure of different tissues and is an indicator of cellularity (12, 41, 42). The apparent diffusion coefficient (ADC) is a parameter derived from DWI, which has been reported to have lower values in higher-grade gliomas (12, 43). Tumor cellularity is inversely correlated with ADC value; decreased ADC values can be seen with processes that degrade cellular integrity such as therapy-induced necrosis or tumor growth (41, 44, 45). ADC has been reported to have the highest values in cystic necrosis, followed by vasogenic edema, and has the lowest values in enhancing tumor (45). Some studies have reported ADC values to be higher in the peritumoral vasogenic edema as opposed to infiltrative peritumoral T2/FLAIR signal, although this has not been a consistent finding across studies (17, 46).

An early prospective study of 17 patients by Asao et al. found significantly lower maximum ADC values in patients who developed tumor recurrence compared to those who had radiation necrosis (45). In a retrospective review of 18 GBM patients who developed enhancing lesions after CRT, Hein et al. found that recurrence and treatment effect could be differentiated with mean ADC values and ADC ratios (ratio of ADC value of the enhancing lesion to the ADC value of the contralateral white matter) with lower ADC and ADC ratios significantly associated with recurrence rather than with treatment effect (41). In a study with MR spectroscopy (MRS) and DWI, Zeng et al. confirmed that ADC values and ADC ratios were significantly higher in radiation injury than in recurrent tumor in 55 patients with high-grade gliomas (HGG) (47). Ellingson et al. have shown that histogram or voxel-based analysis of whole-tumor ADC has a sensitivity of 71% and specificity of 69% and can provide early prognostic info in the form of PFS (48). Gupta et al. conducted a retrospective review of 208 patients which demonstrated that 85% of patients with diffusion restricting lesions developed enhancing tumor at the same site 3 months later, suggesting that low-ADC lesions may precede the development of enhancing tumor (49). In a postmortem study with histopathologic correlation, Nguyen et al. determined the optimal ADC threshold for differentiation of hypercellularity and necrosis as 0.736 × 10⁻³ mm²/s (50).

Tumor heterogeneity in the post-treatment setting complicates the interpretation of ADC values since lesions are generally an admixture of residual/recurrent tumor and necrotic tissue. Several studies showed equivocal results in the diagnostic performance of DWI and ADC in the post-treatment setting (45, 51). Additionally, in a more recent multiparametric investigation by Liu et al., ADC values did not show significant differences between the recurrence and treatment effect groups (52). Another multiparametric 3-Tesla (T) MR approach by Di Costanzo et al. showed that ADC values were higher but not significantly different in radiation injury than recurrent glioma. Prah et al. sought to spatially discriminate between tumor and treatment effect within contrast-enhancing lesions at different stages of treatment and concluded that ADC could not significantly distinguish treatment effect from recurrence whereas perfusion MR parameters of relative cerebral blood volume (rCBV) and normalized cerebral blood flow (nCBF) were able to (53).

The interpretation of DWI and ADC images should be performed alongside non-contrast T1-weighted images and preferably T2^* gradient-echo sequences (GRE) or susceptibility-weighted imaging (SWI) images to avoid false positives in diffusion restricting lesions related to hemorrhage, necrosis, or post-operative tissue injury (54). Smith et al. conducted a prospective study of 44 patients and found that areas of diffusion restriction adjacent to the resection cavity in the immediate post-operative period may resolve and demonstrate subsequent enhancement with eventual encephalomalacia; consequently, they recommend that new enhancement in the immediate post-operative period be interpreted alongside DWI/ADC (55). Mong et al. evaluated persistent diffusion-restricting lesions in patients undergoing BVZ treatment and found that stable diffusion-restricting lesions over time are, in fact, associated with improved outcomes (56).

Although single-shot echo-planar imaging (EPI) is considered the most widely utilized DWI acquisition technique given its advantages of fast acquisition time and availability on clinical MRI scanners, EPI remains sensitive to magnetic field inhomogeneities, resulting in geometric distortions, gradient-induced eddy currents, and T2^*-induced blurring, especially at high spatial resolution and high b-values (57–59). These issues often result from differences in susceptibility between air or venous blood and tissues, as is common in the post-operative setting. Prior solutions have involved using navigator echoes and self-navigated sequences such as periodically rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) and short-axis propeller EPI (SAP-EPI), which reduce sensitivity to magnetic field inhomogeneity and susceptibility artifacts, although these sequences result in increased scan time and thus limited efficiency (59–61). The PROPELLER sequence also corrects for motion degradation as data for each blade can be analyzed to adjust for motion from the patient by oversampling the central k-space with radial sampling resulting in an average of remaining errors (62). Several studies have also demonstrated the advantage of turbo spin echo DWI (TSE-DWI) and readout-segmented echo planar imaging (rsEPI-DWI) in the pediatric brain and at higher magnetic field strengths (58, 62). More recently, Merrem et al. proposed a DWI method which combines a DW spin-echo module with a single-shot stimulated echo acquisition mode MRI (STEAM) sequence and showed that this technique with a modified acquisition and reconstruction strategy results in adequate signal-to-noise ratio (SNR) without susceptibility artifacts and was achievable within 1.5–3 min (57).

In a meta-analysis of 214 patients over 6 studies, Yu et al. noted a pooled sensitivity of 95% with specificity of 83% and supported the auxiliary role of DWI/ADC in the diagnosis of glioma progression (63). Another meta-analysis of 166 patients by Van Dijken et al. noted a pooled sensitivity of 71% and specificity of 87% of ADC values (26) while Du et al. evaluated 17 studies with a total of 656 patients and reported a combined sensitivity of 82% and specificity of 83% (64).

Diffusion Tensor Imaging (DTI) is an extension of DWI and can detect the Brownian motion of water molecules in more than 6 directions (65, 66). DTI is not only used for delineation of white matter tracts for pre-operative guidance but also for assessing the histologic characteristics of tissues and tissue microarchitecture (12, 67). The most common DTI parameters assessed include fractional anisotropy (FA), a measure of the directionality of molecular motion, and mean diffusivity (MD), a measure of the magnitude of diffusion anisotropy (65, 66). Studies have suggested that analysis of FA and ADC values can be used to predict tumor cell density within the core of the tumor or the extent of tumor cell infiltration into the white matter (10, 67, 68). Specifically, the white matter disorganization with tumor infiltration may lead to a decreased FA and increased MD (65). Examples of additional DTI parameters of interest include the axial diffusivity (AD) and radial diffusivity (RD), which refer to the diffusion rate of water parallel and perpendicular to the axon tract, respectively (65).

An early study of 40 patients with intracranial neoplasms including gliomas showed that peritumoral MD and FA values between HGG and low-grade gliomas (LGG) did not differ, but the tumor infiltration index (TII), the difference between expected FA if peritumoral signal were not infiltrated with tumor and the actual FA observed, was significantly different between these two groups (66). However, Kinoshita et al. found that despite the inability of TII to discriminate between vasogenic edema and infiltrative tumor signal, TII correlated with the area of ¹¹C-methionine PET uptake, suggesting its usefulness in delineating the extent of tumor cell invasion into the white matter (67). More recently, a retrospective study of 70 patients with GBM conducted by Bette et al. found that the FA within non-enhancing peritumoral areas were significantly lower in regions that eventually developed tumor recurrence, suggesting that recurrence in the non-enhancing peritumoral region might be predictable by FA metrics at baseline (69). Min et al. found that the regression coefficient of RD to AD (RC_RD−AD) was more effective than FA or TII in distinguishing vasogenic edema from tumor infiltration; when a threshold of 0.6 for RC_RD−AD was set, the sensitivity and specificity of the RC_RD−AD were 85% and 69% (65). They also corroborated the finding that FA of tumor-infiltrated edema was significantly lower than that of pure vasogenic edema although the FA values of the two types of edema had a wide range of overlap (65). In a retrospective study of 41 patients with enhancing lesions within 6 months after chemoradiation, Wang et al. evaluated several DTI and dynamic susceptibility contrast (DSC) parameters, concluding that the best model to distinguish progression from treatment effect employed a combination of the FA, the linear anisotropy coefficient, and the maximum relative cerebral blood volume (rCBV max). DTI is not without its limitations. For example, crossing or coalescing fibers may lead to false negative images (16). Furthermore, cellular infiltration may lead to more diffusion restriction and possibly failure to detect the tract; edema may also artificially increase diffusion and lead to false positives (16).

DSC perfusion imaging measures the susceptibility-induced T2^* signal loss from the injected contrast bolus as it passes through the capillary bed, with the loss of signal depicted as a signal intensity-time curve (12, 17, 74). Subsequently, the AUC is used to derive the relative cerebral blood volume (rCBV), which has become an imaging biomarker for angiogenesis and is widely applied in the diagnosis and treatment monitoring of brain tumors (12, 17, 74). Cerebral blood flow (CBF) can also be easily calculated from the contrast-concentration over time curve (12, 17, 74). rCBV has been used to distinguish higher grade from lower grade gliomas and has been found to be elevated in the peritumoral T2/FLAIR signal of infiltrative glioma as compared to peritumoral vasogenic edema (6, 10, 75). In the post-treatment setting for gliomas, rCBV has been found to be elevated in recurrent or residual tumor more than in PsP or radiation necrosis (50, 76).

Early on, Sugahara et al. demonstrated that normalized rCBV ratios could be used to classify enhancing lesions in treated brain tumors as recurrence or non-neoplastic enhancing tissue, which they confirmed with thallous chloride single-photon emission tomography (²⁰¹Tl-SPECT) (77). Specifically, they noted lesions having a rCBV ratio >2.6 as more likely to be recurrence and lesions with a rCBV ratio <0.6 as more likely to reflect treatment-related changes (77). In a retrospective analysis of 57 post-radiotherapy GBM patients, Barajas et al. found that the rCBV as well as the mean, maximum, and minimum peak height (PH) were significantly higher in patients with recurrent GBM than in patients who had radiation necrosis (78). Boxerman et al. investigated a subset of patients with HGG on a phase 2 clinical trial with TMZ and noted that the change in rCBV at first follow-up visit as well as the overall trend in rCBV rather than the mean rCBV at baseline was significantly different between PsP and progressive disease (79). Mangla et al. also found that the percent change in rCBV at 1 month after CRT with TMZ correlated with survival (80). Blasel et al. found that the maximum rCBV better differentiates tumor progression from recurrence rather than the mean rCBV; they noted that a maximum rCBV of 2.6 had a 78% sensitivity and 86% specificity to detect tumor progression but was not predictive for OS (81). Several studies investigated the diagnostic advantage and clinical value of adding PWI to conventional MR imaging and DWI. Geer et al. reported that with the addition of PWI, confidence in assessment of the post-treatment status increased in 40% of neuroradiologists and 56% of clinicians (82). Kim et al. noted that the addition of DSC (as well as DCE) significantly increased the area under the receiver operating characteristic curve (AUC) for two readers' prediction of recurrent disease (73).

Several larger-scale studies have also shed light on the diagnostic performance of DSC MR in distinguishing recurrence from treatment effect. Snelling et al. reviewed 337 scans from 64 patients and noted the sensitivity and specificity of PWI for HGG as 61 and 88% and for LGG as 86 and 89%, respectively. A meta-analysis including 28 articles by Patel et al. reported a pooled sensitivity of 90% and specificity of 89% for DSC (83). A systematic review and meta-analysis by Wang et al. noted a pooled sensitivity and specificity of 83 and 83% for DSC, respectively (84). In corroboration of these results, a systematic review and meta-analysis by Van Dijken et al. noted a pooled sensitivity and specificity of 87 and 86% for DSC, respectively (26).

One of the disadvantages of DSC is that quantitative rCBV measurements are not feasible given the poor reliability of arterial input functions and unknown voxel-wise contrast agent T2^* relaxivity (85). However, the normalization of rCBV values can provide a means for semiquantification against each patient's own tissue for internal reference. The convention of a user-defined region-of-interest (ROI) within the contralateral normal-appearing parenchyma is susceptible to significant variability in rCBV measurements (85). The use of standardized intensity scales for rCBV maps without the need for user-defined reference ROIs has shown potential for improving the consistency of rCBV measurements (85). Additional limitations of DSC include sensitivity to magnetic field inhomogeneities and to the presence of vascular injury in the post-treatment setting; for example, telangiectasias and aneurysms may increase rCBV while radiotherapy-induced microbleeding in recurrence may result in decreased rCBV (86, 87). A major pitfall of DSC is its underlying assumption that the BBB is intact without contrast leakage or recirculation; if contrast leakage is present, then overshooting may be seen on the DSC time curve (17, 88, 89). Traditionally, a preload bolus technique is used to overcome this, which can be used by itself or in conjunction with DCE (90). For example, a standardized PWI-MR imaging protocol investigated by Anzalone et al. involved splitting a full dose of gadolinium (10 mL gabobutrol)—an initial half-dose (5 mL) bolus was injected 50 s after the start of DCE acquisition and a second half-dose (5 mL) bolus was injected 16 seconds after the start of DSC acquisition (91). Using a full dose for each bolus provides for a higher contrast-to-noise ratio, although a half dose for each bolus with the use of a low flip angle (30 degrees) may provide a good alternative with only a modest decrease in accuracy (90). The preload contrast administration serves to reduce the contaminating T1 effects before the bolus and may also reduce the concentration gradient of contrast extravasation; preload administration is especially recommended for single-echo DSC-MR with high flip angles (91, 92). However, multi-echo instead of single-echo acquisitions and the use of a low flip angle without a preload dose have been proposed for leakage correction (72, 93). Additionally, variation in post-processing methods and software modeling result in inconsistencies in rCBV calculation (88, 94). A recently published set of consensus recommendations for a DSC protocol in HGG by Boxerman et al. is a step toward increased standardization of the DSC technique in post-treatment glioma imaging (90).

More recently, voxel-based, semiquantitative, and quantitative approaches including machine learning techniques have been employed in efforts to achieve a more sensitive and accurate analysis of rCBV. Tsien et al. created parametric response maps (PRM) using DSC parameters such as rCBV to quantify early hemodynamic alterations during treatment and noted a significant difference between the PRM of patients with PsP and those with disease progression (29). In a retrospective study using a semiquantitative histogram analysis of parameters derived from the normalized CBV, Kim et al. showed that the peak height position (PHP) was an independent predictor for the differentiation of recurrence and post-treatment changes with a sensitivity of 90% and specificity of 91% at an optimum threshold of 1.7 while the maximum value (MV) showed a sensitivity of 96.5% and specificity of 93% with an optimum threshold of 2.6 (95). Sanders et al. recently conducted a comparison of synthetically generated DSC-derived rCBV maps from DCE MRI and compared them to the real rCBV maps generated from the acquired DSC MRI and noted a strong correlation between the synthetic and real rCBV maps (74). Thus, the ability to reliably generate a rCBV map from DCE data could potentially spare the patient an additional bolus of gadolinium intravenous (IV) contrast (74).

Gliomas, particularly GBM, are characterized by marked tumor heterogeneity, with areas of solid tumor mixed with necrosis and peritumoral edematous brain parenchyma which is infiltrated by tumor cells (2, 10). Intratumoral heterogeneity is even more important to recognize in the post-treatment setting where the treated lesion is characterized by an admixture of residual tumor cells and necrosis from the cytotoxic effects of CRT (53, 96). Although PWI has great potential in differentiating recurrent tumor vs. treatment effect, its diagnostic performance is limited by the use of gross regional metrics, whereby the mean, median, or maximum values are derived from the entirety of a lesion with abnormal enhancement or T2/FLAIR hyperintensity (53). Consequently, an accurate characterization of the proportion of a treated lesion that is recurrent or residual tumor vs. treatment effect is limited by spatial averaging, which may account for substantial variation in the thresholds for PWI parameters reported in the literature (53). Additional challenges of using PWI-derived metrics include inter-operator subjectivity in the “hot spot” ROI selection for calculation of rCBV values when a ROI is manually selected in an area of tumor with the highest rCBV on a single image (88). However, this method underestimates the entire volume of tumor as well as fails to account for tumor heterogeneity (97). Theoretically, stereotactic tissue sampling throughout a new or worsening area of enhancement of T2/FLAIR signal abnormality would provide insight into how much of a treated lesion consists of residual or recurrent tumor vs. treatment-induced necrosis (53). However, this approach would be prohibitively invasive and unlikely to be feasible in clinical practice. Regional tumor heterogeneity can also thwart diagnostic confirmation via tissue sampling (85). Thus, a non-invasive imaging technique for quantifying tumor heterogeneity has the potential for preventing unnecessary invasive biopsies or repeat resections and more accurately informing clinical decision-making.

The fractional tumor burden (FTB) is an analytical technique that uses PWI data to quantitatively evaluate intralesional heterogeneity in the post-treatment setting. Specifically, FTB is defined as the volumetric fraction of tumor voxels higher than a specified rCBV threshold (97). This method provides perfusion characterization of the entire lesion by using per-voxel measurements of CBV rather than relying on a single value to represent the perfusion of the entire lesion (97). The FTB technique also removes operator dependence with the “hot spot” ROI method which improves reproducibility (72, 97). Another advantage of using FTB is decreased reliance on the magnitude of rCBV values which may be quite variable between institutions secondary to differences in image-acquisition and post-processing techniques (85, 97). Critical to FTB is the use of rCBV thresholds for defining tumor vs. treatment effect.

The groundwork for development of the FTB technique was established by Gasparetto et al. who showed that recurrent tumor and treatment-related necrosis could be discriminated based on whether the rCBV of each lesion was above or below predefined rCBV thresholds (98). In a retrospective analysis of 30 brain tumor patients, they also showed that a unit increase in rCBV increases the probability of recurrence from 57 to 90% and measured an accuracy of 97% in distinguishing enhancing lesions as recurrence or treatment-effect (98). Subsequently, Hu et al. introduced the FTB metric and demonstrated that the FTB of GBM patients with recurrence strongly correlated with histologic tumor fraction as well as OS (96), validating the potential for FTB to serve as an imaging biomarker for tumor progression. They also found that the rCBV mean and mode correlated less strongly with histology compared to FTB and did not correlate with OS (96), which may reflect the advantage of FTB in its ability to account for tumor heterogeneity. Furthermore, Hu et al. noted that a rCBV threshold of 1 differentiates treatment effect and recurrence with 100% accuracy (96). In support of these results, Hoxworth et al. demonstrated that FTB can distinguish between low vs. high histologic tumor content when a set of threshold cutoff values are applied (85). More recently, Iv et al. used preset normalized rCBV thresholds of 1.0 and 1.75 to define low, intermediate, and high FTB in a retrospective evaluation of 47 HGG patients (97). They demonstrated that recurrent tumor exhibited higher FTB and rCBV than treatment effect while treatment effect exhibited lower FTB and rCBV values than recurrence (97). However, intermediate FTB values did not reliably differentiate between recurrence and treatment effect (97). Additionally, there was good consensus agreement among the five physicians involved in this study regarding whether the use of FTB would inform short-term management plans for each case (97). This body of evidence suggests that FTB has the potential to estimate tumor burden accurately and expeditiously during a single MRI study as opposed to obtaining serial ceMR studies for follow-up, which could be used to triage/identify patients who may not need an invasive biopsy or repeat resection.

Figure 2 demonstrates how FTB can be useful as a non-invasive method of depicting tumor heterogeneity and determining if an enhancing lesion consists of more residual/recurrent tumor vs. treatment-induced necrosis. Despite mildly worsening imaging features on conventional MR, visual assessment of the FTB map does not demonstrate an increase in areas of high blood volume within the lesion. In support of this, there is downtrending of the FTB_high percentages on the FTB histograms over several follow-up studies. The FTB findings supported the pathology result of necrosis from resection in this case. FTB also appropriately depicts tumor heterogeneity in Figure 3 where new enhancement and FLAIR hyperintensity is seen in the resection bed 4 months s/p CRT. Although the rCBV and ktrans (see DCE section below) were both increased, the FTB showed a more heterogenous mix of low and high blood volumes throughout the lesion. Pathology results from resection confirmed the lesion as a mix of tumor cells and radiation necrosis.

DCE perfusion imaging is accomplished by consecutively acquiring a series of T1 weighted images before, during, and after IV gadolinium contrast administration to produce a contrast signal time intensity curve (2, 17). The dynamic T1 signal intensity can then be used to characterize the concentration of contrast between the intravascular and extravascular spaces using model-dependent and model-free parameters (2). The volume transfer coefficient, or K^trans is the most commonly calculated parameter and reflects the vascular permeability between the plasma and extracellular space (2, 12, 99). However, K^trans more closely reflects vascular permeability when the BBB is intact; when the BBB is disrupted by tumor, K^trans is more representative of the CBF (17). Evidence in the literature postulates that K^trans is increased in brain tumors likely due to the formation of immature hyperpermeable vessels in the setting of neovascularization; thus, K^trans may be a useful imaging biomarker for tumor proliferation/ recurrence (2, 100–102). Examples of increased K^trans are shown in Figures 3F, 4G. The volume of extravascular extracellular space per unit of volume of tissue, V_e, has also been used to characterize brain tumors (100, 101). Over two studies, Jia et al. demonstrated higher values for K^trans and V_e in HGG than in LGG (100, 101), which reflects increased microvascular permeability of tumor vessels and increased BBB disruption in more aggressive gliomas. Additional studies have shown that K^trans and V_e correlate with the Ki-67 index, a tumor cell proliferation marker that has been useful for glioma grading and prognosis, as well as VEGF expression in gliomas, further supporting the use of these parameters to characterize neovascularization in gliomas (103, 104). Ahn et al. noted that K^trans values were higher in GBM patients with methylated MGMT promoters, which has been associated with PsP and treatment response (105). The V_p, or plasma volume, derived through pharmacokinetic modeling is also a biomarker of tumor neoangiogenesis (86).

In a retrospective study of 37 GBM patients post-CRT with new or increased enhancing lesions, Thomas et al. found that K^trans and V_p were lower in PsP than in disease progression and found that a K^trans of >3.6 had a sensitivity and specificity of 69% and 79% for disease progression (106). Yun et al. also found a lower mean Ktrans in PsP than in disease progression among 33 GBM patients and noted that the mean Ktrans value could predict PsP with 59% sensitivity and 94% specificity (107). Morabito et al. compared the diagnostic accuracy of DSC and DCE MRI in differentiating recurrence from treatment-induced necrosis and concluded that K^trans and rCBV are significantly different between patients with recurrent tumor and radiation necrosis with a sensitivity and specificity of 89 and 97% when an optimal cut-off value of 28.76 for K^trans was used (108). In exploration of a different DCE parameter, Chung et al. used a bimodal histogram analysis of the AUC ratio (ratio of the initial AUC to the final AUC) derived from the DCE signal intensity-time curve to distinguish recurrence from radiation effect with a sensitivity of 94% and specificity of 88% (109). Bisdas et al. investigated several DCE parameters in 18 glioma patients undergoing treatment and found that the Ktrans and AUC were significantly different between recurrent tumor lesions and radiation-induced necrosis (110). In their retrospective study investigating the differentiation of infiltrative tumor from vasogenic edema in the post-BVZ setting, Artzi et al. found that the infiltrative tumor regions were characterized by higher K^trans and V_p than regions of vasogenic edema (6).

More recently, Elshafeey et al. built a machine learning model using radiomic features from K^trans and rCBV maps and were able to achieve a sensitivity of 91%, specificity of 88%, and AUC of 89% for differentiation of PsP from disease progression; in particular, the diagnostic performances of models constructed using the Ktrans and rCBV maps separately were equally high (111). Several meta-analyses report the pooled sensitivities and specificities of DCE MRI for distinguishing recurrence from treatment effect as 92 and 85% (26), 89 and 85% (83), and 73 and 80% (84), respectively.

Compared to DSC MRI, DCE MRI offers better spatial resolution and is not affected by susceptibility artifact; DCE technique also allows absolute quantification of perfusion parameters whereas with DSC technique, CBV values must be normalized to the contralateral normal brain parenchyma (12, 86). However, DCE is somewhat more challenging to implement clinically given the need for pharmacokinetic modeling (21).

Arterial spin labeling (ASL) is a non-contrast MR perfusion technique that uses inversion pulses to magnetically label inflowing arterial blood protons, which then act as natural endogenous tracers for flow rate measurement (17, 112). Thus, ASL is theoretically unaffected by the condition of the BBB, as opposed to conventional ceMR (113, 114). Although ASL imaging does not directly measure the CBV, the CBF can be quantified by comparing the differences in signal between labeled and non-labeled images; studies have shown blood volume and flow are strongly correlated (17, 112, 115). Areas of elevated CBF on ASL images suggest hypervascularity and angiogenesis, which may be predictive of glioma recurrence (17, 116).

Ozsunar et al. compared the diagnostic performance of ASL, DSC, and fluorodeoxyglucose positron emission tomography (FDG-PET) via visual and quantitative analysis in 30 glioma patients and found a sensitivity of 88% for ASL which increased to 94% using a normalized cutoff ratio of 1.3, with lower sensitivities of 81 and 86% for DSC and FDG-PET, respectively (117). Choi et al. applied a grading system to the ASL signal and demonstrated ASL grade to serve as an independent predictor in differentiating PsP from early progression with sensitivity of 79% and specificity of 64% (107). More recently, Manning et al. evaluated the ability of ASL and DSC to differentiate recurrence from PsP and noted that the CBF measured via ASL technique had the highest AUC and misclassified the least number of cases (118). Pellerin et al. reported that compared to 3,4-dihydroxy-6-¹⁸F-fluoro-L-phenylalanine (¹⁸F-DOPA) PET, ASL demonstrated a higher specificity of 100% in tumor isocontour maps whereas ¹⁸F-DOPA-PET demonstrated a higher sensitivity of 94% (119). In a meta-analysis of 20 studies with a total of 939 patients evaluating the value of PWI in the post-treatment glioma setting, the pooled sensitivity and specificity for ASL were 79 and 78%, respectively, with an AUC of 0.88 (84). An additional meta-analysis and systematic review performed by Liu et al. included 10 studies with a total of 368 patients and found that the CBF, rCBF, and rCBV were higher in patients with glioma recurrence than in patients with treatment effect (114).

The advantages of ASL technique include high signal-to-noise ratio (SNR), avoidance of susceptibility artifacts, immediate availability of images, ease of quantification, and lack of IV contrast administration (71). ASL also has better performance around resection cavities as it is less sensitive to prominent surrounding vessels (112, 119). In contrast to DSC technique, the absolute as opposed to the relative CBF values can be measured, which allows for ease of cross-study comparisons and facilitates longitudinal follow-up studies (118). However, ASL is limited by low spatial resolution, lengthened scan time, and motion degradation (118).

Figure 4 demonstrates elevated rCBV, ASL, increased areas of high fractional tumor burden (FTB_high), and vascular permeability which correlate with conventional MR findings of increased enhancement and perilesional FLAIR hyperintensity in a GBM patient with pathology-proven recurrence upon resection.

Single photon emission computed tomography (SPECT) refers to the use of radiopharmaceuticals that localize to areas of tumor with gamma camera imaging. The most common SPECT radiopharmaceuticals used in post-treatment glioma imaging have been ²⁰¹Tl and technetium-99m sestamibi (^99mTc-MIBI) although other less commonly available radiopharmaceuticals such as Tc-99m dimercaptosuccinic acid (^99mTc-DSMA) and Tc-99m glucoheptonate (^99mTc-GHA) have also been employed. The advantages of performing SPECT imaging include the low cost, wider availability, and ease of interpretation, especially with the aid of an accompanying low-dose computed tomography (CT) (128).

²⁰¹Tl is a potassium analog and undergoes active transport through the adenosine triphosphate (ATP) cell membrane transporter; thus, ²⁰¹Tl uptake is related to cellular growth with non-viable tissue exhibiting little ²⁰¹Tl uptake (40, 129). ^99mTc-MIBI passively diffuses through the cell membrane under the transmembrane potential with higher uptake in malignant cells, but its physiological biodistribution in the choroid plexus as well as the temporalis and extraocular muscles limits its use in brain tumor imaging (130, 131). ^99mTc-DSMA is thought to be a phosphorus anion (PO43-) analog with its intracellular accumulation linked to phosphate uptake and kinase pathway activation; its advantage over ^99mTc-MIBI is a lack of uptake across the intact BBB (128). Thus, ^99mTc-MIBI is not taken up in normal brain tissue or choroid plexus (128). ^99mTc glucoheptonate (GHA) is a BBB agent and a structural glucose analog with increased uptake in tumor cells (132). ^99mTc radiopharmaceuticals are generally favored over ²⁰¹Tl currently due to the higher spatial resolution and lower radiation dose achievable with ^99mTc rather than ²⁰¹Tl (40, 128).

In a retrospective review of 19 patients, Tie et al. reported the sensitivity and specificity of ²⁰¹Tl-SPECT for diagnosing recurrence to be 83 and 100%, respectively (129). They observed that the ²⁰¹Tl-SPECT result correctly determined management in 29% and aided the management in 48% of patients and found the diagnostic accuracy of ²⁰¹Tl-SPECT to be superior to conventional MR (129). A study of 201 patients, Le Jeune et al. compared ^99mTc-MIBI-SPECT uptake to stereotactic biopsy results in treated glioma patients and reported the sensitivity, specificity, and accuracy for detecting tumor recurrence as 90, 91.5, and 90.5%, respectively (130). They also observed that a ^99mTc-MIBI-SPECT diagnosis of anaplastic degeneration of LGG was sometimes seen earlier than with conventional MR or clinical characteristics (130). More recently, Roshdy et al. evaluated 30 HGG patients with clinical or radiologic suspicion for recurrence via ^99mTc-MIBI-SPECT/CT and noted that patients with MIBI uptake positive for recurrence were associated with poor survival (131). Amin et al. compared ^99mTc-DSMA-SPECT with MRS for the detection of residual or recurrent disease in 24 glioma patients and found the sensitivity and accuracy for SPECT to be 89 and 92% vs. 61 and 71% for MRS, respectively (128). Santra et al. found the sensitivity, specificity, and accuracy of ^99mTc-GHA-SPECT to be 87%, 97%, and 89% compared to 95%, 24%, and 71% for ceMR, with ^99mTc-GHA-SPECT demonstrating substantially higher specificity and accuracy than ceMR (132). In a recent study, Rani et al. compared the diagnostic performance of ^99mTc-bismethionine-DTPA (MDM) SPECT/CT with DSC MRI for the detection of recurrence and noted a comparable sensitivity and specificity between the two imaging modalities, i.e., 92 and 79% for ^99mTc-MDM SPECT/CT vs. 92 and 71% for DSC MRI (133). Specifically, they noted that the most accurate method for detection of recurrence was a combination of ^99mTc-MDM-SPECT/CT and DSC MR; however, the target to non-target ratio of MDM outperformed normalized CBV in distinguishing between patients with stable vs. progressive disease (133). Nevertheless, the need to perform delayed imaging of 2–4 h after radiotracer injection may be a practical limiting factor for the clinical use of ^99mTc-MDM SPECT/CT (133).

In clinical practice, PET is preferred over SPECT due to its superior spatial resolution which results in increased sensitivity (134). However, SPECT scanners may be more widely available in certain countries and practice settings where access to PET and PET tracers is limited (131).

The use of ¹⁸F-FDG PET to distinguish residual or recurrent tumor from radiation necrosis relies on the concept that malignant cells and treated necrotic tissue have different rates of glucose metabolism. FDG, glucose analog in glycolysis, is trapped in the cell after phosphorylation by hexokinase; thus, FDG uptake is indicative of glucose metabolism (135). Malignant cells generally have high rates of aerobic metabolism while treatment-induced necrosis generally has decreased glucose metabolism (12, 135). Figure 5 demonstrates how FDG PET/CT may be used as a problem-solving tool. A 17-year-old patient with anaplastic astrocytoma post-resection and CRT developed new enhancement, expansile FLAIR hyperintensity, and diffusion restriction in the resection bed with subsequent PET/CT demonstrating hypometabolism, favoring treatment effect rather than recurrence. These suspicious MR findings in the resection bed eventually resolved and were fully appreciated 10 months later.

However, the FDG PET evaluation of brain tumors is limited by high physiologic uptake in normal gray matter. Several studies have reported a lower specificity for FDG PET than other functional and advanced imaging modalities with one study reporting a sensitivity of 81–86% and specificity of 40–94% (135–137). Additional pitfalls of FDG PET imaging are high rates of false positives from inflammatory processes and subclinical seizure activity (129). Furthermore, HGG may not demonstrate high FDG uptake if the degree of necrosis is severe; LGG may be obscured by the high background FDG uptake levels (136).

More recently, the use of PET/MRI combines the excellent soft tissue contrast of MR with the benefit of functional imaging provided by PET tracers (12, 54, 138). Simultaneous acquisition of MR or CT images with PET can also be performed in a reasonably timely manner in the clinical setting. Jena et al. performed a retrospective multiparametric study including the rCBV, mean ADC, Cho/Cr, and the maximum and mean target-to-background FDG uptake ratios and found that the AUC increased from 0.91 to 0.94 when the FDG PET parameter was combined with the mean ADC and Cho/Cr parameters for correctly classifying lesions as recurrence vs. treatment-induced necrosis (54).

Unlike FDG, amino acid (AA) PET is dependent upon the relative discrepancy in intracellular active uptake of amino acid radiopharmaceuticals via the large amino acid transporter system (LAT) between different cell types (139, 140). AA transport is upregulated in tumor cells compared to the normal brain parenchyma with AA transport occurring across intact BBB but low or absent in tissue with treatment-induced injury (54). In contrast to ceMR, BBB disruption is not a prerequisite for intratumoral localization of radiotracer (140, 141). Compared to FDG, AA radiotracers exhibit lower background physiologic uptake in the cerebral cortex and are expected to better differentiate between recurrence and treatment effect (142). Evidence in the literature suggests that the most commonly used AA PET tracers including, ¹¹C-methionine (MET), O-(2-¹⁸F-fluoroethyl)-L-tyrosine (FET), and ¹⁸F-DOPA all demonstrate similar diagnostic performance (143).

¹¹C-MET uptake is associated with increased cellularity as well as microvessel density and is increased in recurrence (113, 139). Deuschl et al. performed ¹¹C-MET PET/MR and conventional MR on a cohort of 50 glioma patients and reported the sensitivities and specificities as 86% and 71% for conventional MRI, 97 and 74% for ¹¹C-MET PET, and 97 and 93% for hybrid ¹¹C-MET PET/MR (144). In a multicenter prospective trial comparing the diagnostic performance of ¹¹C-MET with ¹⁸F-FDG in identifying recurrence, Yamaguchi et al. found that the diagnostic accuracy of ¹¹C-MET-PET at 88% was superior to that of ¹⁸F-FDG-PET at 70% and reported the sensitivities of ¹¹C-MET-PET and ¹⁸F-FDG-PET as 97 and 48%, respectively (142). Deng et al. conducted a meta-analysis of 17 articles and found that ¹¹C-MET PET and DSC-MRI had comparable sensitivities of 87 and 88% as well as specificities of 81 and 85%, respectively (145).

¹⁸F-FET and ¹⁸F-FDOPA, developed subsequently to ¹¹C-MET, have longer half-lives than ¹¹C-MET, meaning they do not require the presence of an on-site cyclotron, which may not be available at all PET imaging centers (139). Bashir et al., among other studies, found a higher ¹⁸F-FET uptake in recurrent lesions and showed that ¹⁸F-FET uptake accurately differentiates treatment effect from recurrence with a sensitivity of 99% and specificity of 94% (146). Similarly, Galldiks et al. found that ¹⁸F-FET uptake is significantly lower in PsP than true progression (147). Puranik et al. found a sensitivity of 80% and specificity of 87.5% for a tumor-to-white matter cutoff value of 2.65 and suggested the use of ¹⁸F-FET PET as a problem-solving tool in the post-treatment setting. Kebir et al. noted that the mean and maximum tumor-to-brain (TBR) ¹⁸F-FET uptake ratios were increased in patients with enhancing lesions which presented 3 months after completion of CRT, suggesting its value in the evaluation of PsP (148). In a recent study by Werner et al., the mean TBR calculated from ¹⁸F-FET-PET showed the highest accuracy among static and dynamic ¹⁸F-FET uptake parameters (149). In the post-BVZ setting, George et al. showed that ¹⁸F-FET uptake decreases after BVZ therapy and only moderately correlates with contrast-enhanced T1 signal intensity; they also found that a high ratio of the ¹⁸F-FET uptake before and after treatment as well as an increased correlation between ¹⁸F-FET uptake and contrast-enhancing signal were associated with poor PFS and OS (150). The advantages of ¹⁸F-FET-PET include the lack of physiologic uptake in the striatum and relative ease and lower cost of synthesis (140, 151), whereas its limitations include non-specific tracer uptake secondary to slow renal elimination, lack of protocol standardization, and the potential need for delayed imaging (140).

Like ¹⁸F-FET, ¹⁸F-FDOPA is also transported across intact BBB and detects the extent of tumor beyond contrast enhancement in conventional MR (54). ¹⁸F-FDOPA is known to have a higher sensitivity for evaluation of LGG (54). Fraioli et al. demonstrated that the volume of residual tumor as characterized by ¹⁸F-FDOPA-PET is larger for HGG than for LGG compared with the tumor extent as depicted by conventional MR; ¹⁸F-FDOPA-PET was also able to detect recurrent disease in non-enhancing regions on ceMR (141). Herrmann et al. demonstrated that both visual assessment and semiquantitative assessment of ¹⁸F-FDOPA uptake with the mean and maximum standardized uptake values (SUV_mean and SUV_max) values were highly accurate (82% for visual and 77–82% for semiquantitative assessment) and were significant predictors of PFS (152). Karunanithi et al. showed that in comparison to ¹⁸F-FDG, ¹⁸F-FDOPA performed higher in sensitivity and accuracy but not specificity (136). They reported the sensitivity, specificity, and accuracy for detecting tumor recurrence as 48, 100, and 61% for ¹⁸F-FDG-PET/CT and 100, 86, and 96% for ¹⁸F-FDOPA-PET/CT, respectively (136). Jena et al. found that when advanced MR parameters from ceMR, DSC-MRI, DWI, and MRS were interpreted in conjunction with ¹⁸F-FDOPA-PET, the diagnostic performance of the visual qualitative assessment of recurrence achieved an accuracy, sensitivity, and specificity of 95%, 96%, and nearly 100% (54). Youland et al. also confirmed that when combined with ceMR, ¹⁸F-FDOPA-PET improved the sensitivity and specificity to 94 and 75% (137). A particular advantage of ¹⁸F-FDOPA-PET is the rapid scan time with uptake levels peaking between 10 and 30 min after injection while ¹⁸F-FET-PET may require delayed phase imaging, which may be challenging for patients who have difficulty tolerating the exam (141). However, ¹⁸F-FDOPA-PET demonstrates physiologic uptake in the corpus striatum which could potentially obscure lesions close to the basal ganglia (153). Furthermore, the radiotracer is difficult to synthesize and may be limited in its availability (143). In a systematic review of the overall utility of AA PET in the post-BVZ assessment of treatment response, Hughes et al. found that AA PET is particularly useful in monitoring treatment response to BVZ in HGG patients (139). Additionally, ¹⁸F-FET-PET was able to diagnose BVZ treatment failure earlier than conventional MR (139).