Performance of amide proton transfer imaging to differentiate true progression from therapy-related changes in gliomas and metastases

This review identified substantial potential for APT-CEST to improve the non-invasive discrimination of TRC from TP/TR in gliomas after CRT. Moreover, based on the same-population analysis, this study delivers limited evidence that APT-CEST performs similarly to PWI and better than conventional MRI, DWI, DTI and MRS for this clinical question. Combining APT-CEST with other imaging techniques in gliomas can improve diagnostic accuracy in a potential clinical setting. For BM, the level of evidence is substantially weaker due to limited data and less unequivocal outcomes.

The findings of this study corroborate preliminary findings from an earlier review of gliomas [12]. The theoretical backdrop is that the APT signal is relatively higher in TP/TR due to increased cellularity and protein content in tumour tissue compared to RT-induced coagulative necrosis and loss of cytosolic components in TRC. As intracellular cytosolic proteins are a potential source of APT signal, loss of these components could lead to decreased saturation transfer from amide to water and subsequent signal decrease. APT signal increase in large cysts or haemorrhage may limit accuracy but can be corrected through co-registration on conventional MRI [19], further increasing diagnostic performance [21] (see Fig. 6). Moreover, APT combined with PWI allows for assessing two distinct tumour characteristics, protein content and neovascularisation, which could work synergistically and correct their individual limitations to provide higher diagnostic performance. However, this review did not identify a discriminatory capability in BM using MTR_{asym(3.5 ppm).} In contrast, PWI, PET, and conventional MRI all demonstrated some capability to differentiate TP/TR from TRC in BM [33]. Peculiarly, both BM APT-CEST articles included the highest TRC percentage among all studies, which is attributable to stereotactic radiosurgery in BM, causing a high incidence of RN. This could explain why MTR_{asym(3.5 ppm)} imaging performed poorly compared to glioma. The authors postulate that the subtraction of MTR_Amide from MTR_NOE, both of which did individually provide significant discriminatory capacities, could diminish the signal difference between TP/TR and TRC. One metastasis study [29] also exclusively used a non-standard saturation pulse amplitude (0.52 μT), resulting in a large NOE contribution and low MTR_{asym(3.5 ppm)} signal [34]. Another explanation for the lack of signal difference in BM could include an a priori lower signal intensity in metastases compared to primary brain tumours [35]. Furthermore, both studies included various types of BM, possibly confounding APT signals as they differ in histological characteristics. Unfortunately, literature on the APT signal in different BM types is lacking. The underrepresentation of BM in this study, and APT-CEST research in general, makes drawing any conclusions about the clinical applicability challenging. Additional research is required, using various acquisition parameters and processing metrics, to determine the utility of APT-CEST in treatment response assessment in metastasis patients treated with stereotactic radiosurgery in a clinical setting. Kroh et al [26] was the only glioma study where APT performed insufficiently. Z-spectral fitting-based and relaxation-compensated approaches did not provide significant diagnostic performance either. The latter phenomenon can be explained through a large saturation bandwidth at 2 μT amplitude at 3 T, which can make APT signal invisible [10]. NOE, direct effect, and amide signals all contribute, in varying degrees, to the APT-weighted signal. The amount of APT-weighted signal in TP/TR that is a result of pure amide proton transfer, is a matter yet to be elucidated. In addition, there are T1-contributions and spill-over effects, the latter of which could be corrected for through a relaxation-compensated approach. More research is required to validate the utility of Z-spectral fitting-based and relaxation-compensated approaches in APT imaging in post-treatment response assessment.

Some limitations apply to the articles included in this review and meta-analysis. First, the effect of irradiation dose, which possibly increases RN incidence, on APT signal could not be investigated due to unreported doses in most studies. Additionally, methylguanine-DNA methyltransferase (MGMT) methylation might influence the incidence of PsP in glioblastomas [36]. Unfortunately, the included studies either did not report or separately analyse different tumour mutations. Additionally, the fact that most studies did not exclusively apply histopathological verification as a reference standard, and used outdated WHO criteria for glioma grading raises concerns regarding the clinical applicability of these results. Furthermore, some studies assessed both lower and high-grade glioma (LGG/HGG), possibly confounding APT signal as recent literature suggests that APT-weighted signal differs between LGG and HGG [37]. However, glioma subtypes did not appear to cause significant heterogeneity in this meta-analysis, which could result from the small number of patients with LGG compared to HGG. To account for threshold effect-induced heterogeneity, this review provides an overview of the diagnostic performance of APT-CEST at various imaging parameters with their respective optimal thresholds. Future research can focus on validating and/or implementing these parameters and thresholds to further explore and facilitate clinical implementation. Additionally, this review has several limitations. First, this review and meta-analysis only includes a small number of studies because of the novelty of and the limited literature regarding this subject. Taking the limited data and significant inter-study heterogeneity at the outcome level into account, the pooled forest plot results were omitted. While the meta-regression and subgroup analysis may hint at possible biases and/or heterogeneity-inducing factors in the data, the interpretation of these results is challenging considering the low sample sizes and low statistical power resulting from a limited number of studies. Another limitation is that the included studies assessed patients within 2 years after radiotherapy completion to include both PsP and RN, which both classify as TRC. Unfortunately, this leads to a large variation in the interval between therapy and APT imaging across studies. Furthermore, some studies included outliers, with patients receiving APT imaging up to 10 years [21] after completion of radiotherapy. Moreover, the gold standard for differentiation between TP/TR and TRC is histopathological assessment after biopsy or resection. It was thus regarded as the optimal reference standard in this review, but it is still prone to sampling bias [38]. Furthermore, two studies [18, 21] had (partially) overlapping cohorts with two other studies [25, 27]. However, both studies were excluded from the meta-analysis to avoid cohort overlap. The number of included studies comparing APT-CEST to other imaging techniques is small. For this reason, pooled results of comparative studies were reported separately. The forest plot analysis, however, provides a visual indicator for future comparable studies regarding the potentially achievable sensitivities and specificities, which for glioma range close to those of other quantitative MRI techniques, hinting at the clinical utility of APT-CEST [11]. Additionally, some aspects need to be noted in this review. Although Z-spectral fitting-based and relaxation-compensated approaches did perform well in BM [28, 29], only results of MTR_asym(3.5 ppm) APT at 3 T were included in the meta-analysis to facilitate the harmonisation of efforts in future research, as MTR_asym(3.5 ppm) at 3 T is most commonly used for brain tumour assessment in a clinical setting [10]. While the MRI parameters for APT image acquisition varied, they did not cause any statistical heterogeneity after the removal of BM studies. Furthermore, the risk of bias assessment in this review was quite strict to better represent the real-world applicability of APT-CEST for treatment response assessment. Finally, publication bias was not assessed in this review due to the limited number of studies included for analysis. Additionally, funnel plots of the diagnostic odds ratio bode low statistical power and are difficult to interpret if multiple index tests are being evaluated [39].

In conclusion, APT-CEST has shown significant promise in discriminating TP/TR from TRC in post-treatment gliomas. However, this has not been demonstrated in metastases. In addition, APT-CEST imaging has shown limited evidence for superior diagnostic performance to other conventional or advanced MRI techniques. A combination of APT-CEST with PWI and/or conventional MRI techniques appears to provide the best diagnostic performance, outperforming each single imaging modality. As such, APT-CEST has demonstrated potential as an alternative to biopsy in post-treatment response assessment of gliomas in a clinical setting. Further validation of these results, as well as a higher degree of inter-centre technique harmonisation, are warranted for future research.