Qualitative and quantitative DTI image analysis, as presented in this study, has clearly shown the distinctive imaging features of radiation-induced necrosis and viable glioma in rat models. Moreover, there is a close correlation between the DTI indices and the histological characteristics for these two pathologies.
It has been recognized that organized microstructures with high FA exist in viable glioma, where the brain parenchyma has been destroyed, due to tumor cells proliferating rapidly and packing closely in a coherent way. This is supported by some early pathology experiments in humans [26
] as well as several recent DTI observations, both in animals [17
] and in humans [27
]. Interestingly, the FA maps reveal specific tumor cell arrangements within the peripheral zones of 9L and GBM22 xenografts (circular versus radial, respectively). On the other hand, radiation necrosis is associated with random microstructures, which reflects brain tissue necrosis and reactive reflection. Our results imply that FA has the potential to differentiate radiation necrosis from glioma with higher diffusion anisotropy.
The ADC is another important DTI index to measure the magnitude of water diffusion. Our preclinical results have indicated that the central zone of radiation necrosis is slightly hypointense ( and ). The ADC values in the central zone were significantly lower in radiation necrosis than in either 9L gliosarcoma (with high density of tumor cells) or GBM22 (with spontaneous necrosis). This is because coagulative necrosis with fibrosis deposit () is a major pathological change in the central zone of radiation necrosis. In this condition, the diffusion of water molecules is restricted in all directions due to decreased water content and obstacles from fibrosis deposit. On the contrary, the necrotic periphery shows a hyperintensity on the ADC map, which is associated with various complicated pathological changes, such as endothelial thickening, vascular dilation, and vasogenic edema [15
]. Unlike radiation necrosis, there are consistently high ADC values in the entire glioma regions (), which have been attributed to the increase in perivascular space and micro-necrosis in the tumor [31
can provide additional information about the directionality of water diffusion. These indices have been used to characterize myelin and axonal damage in white matter diseases [31
]. In radiation necrosis, λ//
were significantly decreased in the central zone, compared to the peripheral zone, but the FA values were not different for these two regions. Our ROC analysis results showed that λ//
in the central zone and peripheral zone, as well as λ
in the central zone, all provide satisfactory diagnostic powers to classify radiation necrosis or gliomas. Their diagnostic powers are approximately equal to ADC’s and higher than FA’s. It seems that, rather than the FA, the use of the ADC, λ//
, and λ
, is better for distinguishing between radiation necrosis and glioma, although this needs to be validated further in patients who have more complex and histopathologically heterogeneous lesions.
When comparing radiation necrosis with glioma, it should be noted that the presence of spontaneous necrosis, commonly observed in high-grade gliomas, may be a confounder for radiation necrosis, because spontaneous necrosis would also be associated with low FA values, as observed in the central zone of the GBM22 tumor. Our results have shown that there are also different diffusion characteristics for the 9L and the GBM22 tumors, probably because of the spontaneous necrosis that leads to a very low FA value in the GBM22 central zone.
It is important to note that there are currently contradictory reports on the differentiation of radiation necrosis from tumor recurrence in patients using the DTI indices. For example, Sundgren et al. reported significantly decreased ADC, λ//
, and λ
values in radiation necrosis, compared to tumor recurrence, while no difference was found for FA [31
]. However, several other researchers reported increased ADC and decreased FA values in radiation necrosis, compared to tumor recurrence [33
]. This disagreement may be attributed to several factors. First, the quantitative analysis in the clinic is usually done in new contrast-enhancing regions, where delayed radiation injury, irradiated tumor, and tumor recurrence may coexist. The disruption of the blood-brain barrier for these pathologies is indistinguishable using standard MRI. Moreover, it is always very difficult to have the exact correspondence between the ROIs for the DTI analysis and tissue sampling for pathology. Based on our findings (), delayed radiation-induced necrosis consists of an ADC-hypointense central zone and an ADC-hyperintense peripheral zone, and increased and decreased ADC values can both be observed in the lesion, depending on the localization of the ROIs. In addition, it is know that the ADC values in brain tumors increase following treatment and drop to baseline when the tumors re-grow [36
]. Therefore, the clinic results may differ between labs, as pointed out previously [31
]. Second, clinical studies have reported that FA values differ within GBM, anaplastic astrocytoma, or pilocytic astrocytoma [12
]. In particular, significant differences in FA were observed between grade-1 (mean FA: 0.150 ± 0.017) or grade-2 (0.159 ± 0.018), and grade-3 (0.230 ± 0.033) or grade-4 (0.229 ± 0.033) gliomas [37
]. Obviously, when these gliomas consisting of various grades (thus various DTI indices) are compared as a whole with radiation necrosis, the predictive powers of the DTI indices in the differentiation between these two lesions will be reduced, and the results will be varied in different labs.
Care should be taken when translating our pre-clinical results to clinical practices. First, in the current pre-clinical study, we compared “pure” radiation necrosis with “pure” gliomas and further divided the lesions into the central and peripheral regions. However, "pure" radiation necrosis does not typically manifest itself in patients with gliomas because there is always an element of tumor cells left behind. Additionally, the relevant clinical question is typically "tumor recurrence" and not "pure" untreated gliomas. Moreover, radiation necrosis and recurrent tumor may often co-exist in patients. Second, radiation necrosis and gliomas are more heterogeneous in clinical conditions than in animal models. The strict periphery and central region observed in radiation necrosis are likely due to the well-controlled, focused radiation field in the current study. These characteristics are consistent with some previous reports of human cases [31
], but not always observed in lesions thought to be radiation necrosis [33
]. In spite of such complicity in real clinical situations, our pre-clinical results could provide a clear, reliable reference for numerous ongoing investigations into DTI’s ability to assess more heterogeneous radiation necrosis and gliomas in humans. In addition, our findings would be useful for the diagnosis of radiation-induced white matter necrosis in patients with nasopharyngeal carcinoma who have received radiation therapy [38
Finally, there are several technical limitations in this study. A limitation is that the diffusion-weighted images were acquired only in six gradient directions. It is known that errors in FA decreased with the number of diffusion gradient directions [39
], so six was relatively small. Another one is that we reconstructed DTI maps by averaging the images over the different echo times to improve the signal-to-noise ratio [41
]. However, each echo technically has slightly different diffusion weighting due to the small extra contributions from crushers around 180°, read gradients, etc. These limitations may affect the precision and accuracy of DTI measurements.
In conclusion, our study suggests that delayed radiation-induced necrosis and viable glioma in rat models exhibit different DTI features. Particularly, the magnitudes and directionality of water diffusion are decreased in the central zone of radiation necrosis. DTI indices provide useful diagnostic information to distinguish between these two pathologies. These results may influence the clinical formulation of a treatment plan in the clinic.