The current systematic review and meta-analysis aimed to compare the most promising methods of the differentiation of PsP and TTP in patients with high-grade gliomas. A prior meta-analysis has compared the utility of DWI and PWI for discriminating PsP and TTP.[33] Consistent with our results, they did not find a significant difference between the two modalities. In contrast to the study by Tsakiris and colleagues, our inclusion criteria were not limited to two methods only. Furthermore, we consider each modality in the context of its clinical utility, aiming to provide a recommendation for physicians.
A total of seven PsP and TTP imaging discriminators have been identified: 1) DSC-pMRI, 2) DWI, 3) DCE-pMRI, 4) ASL, 5) APTw-MRI, 6) 18F-FET PET, and 7) conventional MRI. Two combinations of the methods were included: 1) DSC-pMRI & DCE-pMRI, and 2) DSC-pMRI & ASL.
In the current review, the diagnostic accuracy measures, including the sensitivity and specificity for each method were extracted from the included studies and summarised in a forest plot. [34] Sensitivity was defined as the proportion of patients with histologically confirmed pseudoprogression out of all patients with suspected pseudoprogression based on neuroimaging results. Specificity is the ability of the test to identify cases of true tumour progression (and exclude pseudoprogression). It was defined as the proportion of patients with confirmed true tumour progression out of all patients that were deemed to have tumour progression based on neuroimaging findings. A high sensitivity and specificity constitute a high diagnostic accuracy, which in turn influences increases the overall precision of clinical decisions.
Diffusion and Perfusion-based Methods
Our meta-analysis found that DWI, DSC-pMRI and DCE-pMRI have high potential for differentiating PsP from TTP in patients harbouring HGGs. Out of these three methods, DWI demonstrated the highest sensitivity for detecting TTP (0.91 [0.84 – 0.95]), and DSC-pMRI demonstrated the highest specificity (0.89 [0.68 – 0.97]). A comparison of sensitivity and specificity for DWI, DSC-pMRI and DCE-pMRI did not reveal any statistically significant differences.
Assessment of imaging results using pre-specified parameter cut-off values was associated with higher sensitivity and specificity values in comparison to studies that relied on visual inspection. Kerkhof and colleagues[31] visually inspected rCBV colour maps to differentiate PsP from TTP, which yielded 72% sensitivity and 23% specificity, both of which were the lowest values in the DSC-pMRI subgroup. In contrast, two other included DSC-pMRI studies[35],[26] reported 100% sensitivity and 100% specificity using parameter cut-off values to differentiate PsP from TTP. Jovanovic and colleagues used a ratio of 2.89 of normalized CBV between the lesion and normal-appearing tissue, while Martínez-Martínez and colleagues used an rCBV value of 0.9.
A direct comparison of DSC-pMRI and DWI was provided in studies by Kim and colleagues[24] and Prager and colleagues[25]. The former group found that the maximum CBV parameter of DSC-pMRI and the mean apparent diffusion coefficient (ADC) of DWI differentiated PsP from TTP with the same sensitivity (79%) and specificity (45%) in 34 patients. Prager and colleagues found DSC-pMRI to be 100% sensitive and 75% specific, although there were only four cases of PsP with available DSC-pMRI data. DWI performed with a sensitivity of 95% and specificity of 63% in the same study with eight PsP-confirmed patients.
Choi and colleagues[27] investigated the diagnostic accuracy of DSC-pMRI and ASL. The sensitivity and specificity of DSC-pMRI were determined as 82.4% and 67.9%, respectively and 79.4% and 64.3%, respectively, for ASL. A combination of the two modalities resulted in an increased sensitivity and specificity of 94.1% and 82.1%, although this did not represent a significant increase in diagnostic accuracy (p = 0.133). Jovanovic and colleagues[26] separately assessed DSC-pMRI and ASL, and quantitative analysis found both methods yielded 100% sensitivity in their patient sample. For specificity, ASL scored 73% compared to 100% for DSC-pMRI.[26] All four included diffusion/perfusion-based methods show clinical potential. DSC-pMRI is currently the most widely employed, and its protocol and acquisition parameters are already well-defined.[36]
FET PET
There has been increasing interest in the application of PET in differentiating between PsP and TTP. One study included in this meta-analysis used 18F-FET PET and found the maximal tumour-to-brain ratio (TBRmax) differentiates between the two with 100% specificity and 91% sensitivity at a cut-off of 2.3, in a sample of 22 patients.[28] A similar cut-off value of TBRmax = 2.55 was reported by Kebir and colleagues[37]. In the same study, a linear discriminant analysis-based algorithm was trained on IDH-wildtype glioblastoma FET PET features and compared the results to a conventional FET PET analysis.[37] The algorithm provided an AUC of 0.93, which was higher than the AUC for TBRmax of 0.68. Although this difference was significant when compared to other included parameters, it was not compared to TBRmax (p = 0.081).[37] This study was limited by a sample of 44 patients,[37] so further research with larger datasets would be necessary before considering clinical implementation.
APTw-MRI
APTw-MRI was used to differentiate PsP and TTP in one study only. Ma and colleagues[29] found APTw-MRI to correctly identify 19 out of 20 patients in their TTP cohort (95%) and 11 out of 12 patients in their PsP cohort (92%). There was a marked signal increase in the TTP compared to PsP cohort, with an APTWmean cut-off of 2.42% and an APTWmax cut-off of 2.54%.
Combination Methods
Multimodal approaches often demonstrate increased diagnostic accuracy and provide an additional layer of confidence compared to individual modalities. It is reasonable to assume the highest diagnostic accuracy would be achieved from the combination of results from already established modalities. However, the trade-off is the accompanied increase in cost and acquisition time. Regardless, with increasing availability of several above-mentioned modalities, the advantage of combination methods should be considered.
Two combination methods were included in the present review. A combination of Ktrans and rCBV maps obtained from DCE-pMRI and DSC-pMRI acquisitions, respectively, reported high sensitivity (88%) and specificity (91%) when applied to a cohort of 98 patients.[38] The maps could not discriminate between PsP and TTP in the cohort when used individually. Choi and colleagues[27] combined DSC-pMRI with ASL and reported sensitivity and specificity of 94% and 82%, respectively, also finding the combination values higher than the individual methods. A prospective trial is currently registered to assess the diagnostic accuracy of DWI and ASL.[39]
Clinical Utility
Despite the large number of studies reporting the diagnostic potential of different imaging protocols, their routine clinical use has not been implemented. A summary of the main clinically relevant parameters is presented in Table 3. [40],[41],[42]
An inherent limitation of using perfusion-weighted imaging is that while perfusion parameters are generally lower in PsP, the associated inflammatory response is likely to influence perfusion and lead to increased perfusion parameters such as rCBV.[43] Similar effects have been seen with DWI as a result of radiation necrosis, suggesting decreased ADC may not always reflect a high cellularity and TTP.[44] However, both PWI and DWI appear to demonstrate overall high diagnostic accuracy.
As the most frequently used perfusion MRI modality, DSC-pMRI may be preferable for standard protocol due to its high clinical availability and short acquisition time that can be under one minute.[42] The standardisation of rCBV discriminating cut-off values is limited by numerous potential imaging and data processing artifacts impeding accurate perfusion quantification as outlined by Willats and Calamante’s 39 steps for accurate perfusion of DSC-pMRI data.[45] One of the most widely discussed issues is the possibility of contrast agent leakage into extracellular tissue, known as T1 shine-through effect [46]. Application of model-based leakage corrections is advised for single-echo gadolinium-based DSC-pMRI to account for the extent of vascular permeability.[47]
DCE-pMRI has a high signal-to-noise ratio compared to the other MR-perfusion techniques.[48] The main limitation of this method is the relatively long data acquisition time, often over several minutes. [49] Similar to other perfusion techniques, full quantification remains challenging due to difficulty in the DCE tracer modelling. Efforts are currently undertaken to resolve issues related to accurate quantification of perfusion techniques. The establishment of taskforces such as the Quantitative Imaging Biomarkers alliance will facilitate clinical implementation of methods by providing reference measures and guidelines for best practices [50].
ASL was a less frequently reported discriminating method compared to other perfusion methods. The main advantage of ASL over DSC-pMRI is that it does not require a gadolinium-based bolus injection. It may therefore be more suitable for patients with contraindications to administration of contrast agents.[51] Furthermore, ASL can acquire entirely quantitative values of cerebral blood flow (CBF). A non-significant increase in sensitivity and specificity was observed when CBF measures acquired using ASL was combined with DSC-pMRI, compared to use of the methods.[27] Jovanovic and colleagues[26] concluded that the diagnostic accuracy of ASL was sufficient to replace DSC-pMRI and therefore, avoid repeat follow-up contrast injections.[26] An important consideration of ASL is the longer acquisition time of 8-10 minutes at 1.5T and 4-5 minutes at 3T as well as significantly lower signal-to-noise ratio (SNR) compared to other perfusion methods.[42]
APTw is a novel imaging technique demonstrated to detect the increased mobile protein content in brain tumours[52]. Its full potential is yet to be established as U.S. Food and Drug Administration (FDA) approval of 3D-APTw for use on 3T clinical MRI scanners was granted in 2018.[53] However, APTw examinations may be time consuming (~5-10 minutes) and are susceptible to magnetic field inhomogeneities.[54] Some recent work aims to optimise the signal-to-noise ratio[55] and image acquisition speed. APTw is a promising method with initial studies reporting a high diagnostic accuracy, but larger datasets are needed to compare its performance against other techniques.
Lastly, one study investigated the use of 18F-FET PET in diagnosing PsP.[28] Despite its high sensitivity and specificity, a long acquisition time of 50 minutes as reported by Galldiks and colleagues[28] limits clinical potential. Since 18F-FET PET relies on administration of labelled amino acid analogue, patients in the study were also required to fast for at least 12 hours before scanning. In contrast to other radiotracers, the half-life of fluorine-18 is long enough to allow for off-site production. The requirement for pharmacokinetic analysis with compartment modelling[56] further limits potential for clinical implementation.
Future Directions
Quantitative methods offer a more objective approach towards finding patterns in clinical data and enable more accurate diagnosis compared to qualitative methods.[57],[58]
Jang and colleagues[59] recently applied a deep learning approach using convolutional neural networks to the differentiation of pseudoprogression and true progression, and achieved a sensitivity of 87% and a specificity of 94.5%. Another study found a benefit of the combination of hypervascularity, cellularity and permeability parameters over single parameter measurements to distinguish the conditions.[60] The need for large datasets for training and testing radiomics models has led to a general lack of power, therefore future research should focus on increasing accessibility and data availability. National support for the scaling of technology and the potential use of artificial intelligence to aid clinical decision making has been outlined in the NHS Long Term Plan.[61]