The continuous technical development in MRI with fast imaging techniques, resulting in protocols with scan times between 15–30 minutes [15], has made chest MRI an interesting alternative to CT [16, 17]. Regarding the low evidence of appropriate follow-up examinations in NSCLC, especially after chemoradiation [8], chest MRI might be the future solution. Our study is one of the first longitudinal investigations comparing CT and MRI after chemoradation of NSCLC regarding morphological and functional parameters.
Prior studies have mainly focused on pretreatment comparison of CT or PET/CT and MRI [15, 18]. Fleckenstein et al. already demonstrated a high level of concordance between the pretreatment tumor volumes of PET-CT and DWI for radiotherapy-planning [18].
FDG-PET/CT has a high diagnostic accuracy in the detection of local tumor recurrences[19] and is often used, when – based on the follow-up CT – a recurrence is suspected[20]. Nevertheless, FDG-uptake is also enhanced in lung regions which show severe RILT. Therefore, the diagnosis of local lung recurrences in areas of RILT might be impaired in such cases. Also, FDG-PET/CT is usually more expensive and less available than DWI and thus contraindicated as a routine measure in the follow-up due to economic and logistic reasons.
We observed no statistically significant difference regarding the percentage change of the longest longitudinal diameter between CT and DWI at 3, 6 and 12 months (Fig. 1). The spread of the measurements within the two modalities can be explained most easily on the different initial tumor sizes and its responses. An inverse tumor response between CT and DWI was not observed. This is also reflected by the classification of the tumor response according to RECIST 1.1. There was a high concordance between DWI and CT regarding therapy response after 3 months resulting in PR (n = 6), SD (n = 3) and PD (n = 1). In the three discrepant cases only a few percentage points were missing for the transition from PR to SD or SD to PR. However, at 6 and 12 months, there is a diminishing correspondance of CT and DWI in terms of tumor response. This observation may have multifactorial causes, such as RILT with limited sensitivity of CT scans, deceased patients and missed appointments. However, in the follow-up at 6 and 12 months in individual cases, the DWI detected recurrences earlier than CT or excluded them with greater certainty (Table 1[Cases 2 and 7] and Fig. 2). These results are in line with the study published in May 2019 by Usuda et al. [21]. In their study DWI was more accurate than CT in determining a response of recurrent lesions of lung cancer to chemotherapy and/or radiotherapy. Consistent with our study, they concluded that DWI may be able to identify residual cancer, thereby improving specificity and sensitivity.
Conventional response criteria like RECIST 1.1 have some limitations. There is an ongoing debate how accurate a unidimensional measurement can represent the real tumor burden due to varying and often highly irregular tumor shapes. Meanwhile, several studies have demonstrated that volume measurement in lung tumors is more reproducible than size measurement [22, 23]. In addition, the study by Zhang et al. proves that DWI has a more precise delineation of lung cancer while exhibiting higher reproducibility [24]. In our study there was no significant difference between tumor volumes as determined by CT and DWI at any of the three follow-up dates. The tumor volumes in the DWI tended to be slightly smaller than in their CT counterparts. We identified the more precise demarcation of the tumor against atelectatic lung tissue and parenchymal changes in pneumonitis as the major cause for this discrepancy. There is a lack of data comparing CT and DWI tumor volumes in the course of therapy after chemoradiation. A comparable study by Weiss et al. determined significantly larger tumor volumes by CT as compared to DWI in patients after chemoradiation [25]. However, the results are only partially comparable to the data presented here, since the working group around Weiss et. al chose follow up assessments at 3 and 6 weeks, thus focusing on early changes.
In addition to assessing tumor response, the applied imaging modality should reliably indicate RILT. RILT typically occurs as early as 4 to 12 weeks after treatment and may transform into radiation fibrosis (which may also occur independently) after 6 months or later [26]. Although clinically debilitating pneumonitis (grade ≥ 3) after radiotherapy develops in less than ten percent of patients [27], imaging in commonly used scores, such as the LENT-SOMA Score from the European Organization for Research and Treatment of Cancer (EORTC) [28], plays a role in diagnosis and therefore therapy. Regarding the evaluation of RILT, some groups use functional investigations in MRI with xenon gas with quite impressive results [29]. Meanwhile, the parenchymal structure of the lungs can be adequately assessed by MRI, as was shown by Sileo et al. in patients suffering from cystic fibrosis [30]. To our knowledge, we hereby present the first investigation to examine the correlation of RILT scores determined by CT and MRI, respectively respiratory gated T2-weighted sequence, over a period of one year. At 3 and 12 months a high correlation of the RILT scores was found. However, at the early stage of fibrosis development at 6 months, only a moderate correlation was shown between the two modalities. Hence, respiratory gated T2-weighted sequence can adequately assess the ultrastructure of the lungs in the early and late phase after chemoradiation to diagnose or exclude RILT. Differentiating between treatment effects like RILT or tumor-atelectasis-complex and residual or recurrent tumor, is challenging [31]. Like aforementioned, in some cases of the presented group, by using DWI, as compared to CT, we were able to delineate recurrences earlier and to more reliably rule out recurrence within lung parenchyma altered by RILT (Fig. 2). These results are in line with a study of Munoz-Schuffenegger et al. in which they could prove that DWI confirmed the suspicion of local recurrence in patients with highly suspicious CT scans [32]. Furthermore, DWI/ADC not only provides these important additional informations but might also be a prognostic factor.
Looking at the individual mean ADC values of patients over the time course, no clear pattern could be observed in our study (Fig. 4A). However, averaging the ADC values of all patients at each time point mean and maximum ADC showed a tendency to increase (Fig. 4C/D). Our results are consistent with prior studies which demonstrated a significant ADC increase after chemoradiation and chemotherapy [25, 33, 34]. Weiss et al. showed that patients with survival < 12 months had a lower increase in ADC values compared to longer-lived patients [25]. Sampath et al. could demonstrate that an ADC increase of 40% at 1 month after SBRT for NSCLC is associated with a higher rate of local failure [35]. In contrast, non-responders in the study by Chang et al. had a slight decrease in ADC, whereas responders had a relatively steeper increase of ADC [33]. As opposed to the latter data, after formation of a PD and PR group, we were unable to detect a significant increase or decrease in the mean ADC between the two groups (Fig. 4B), which could be due to the small sample size. However, both the pretherapeutic and the mean ADC values over the course tend to be lower in non-responders (PD group). In agreement with our findings, Shintani et al. and Iizuka et al. found that low ADC on pre-treatment MRI were associated with local recurrence and poor disease progression [36, 37]. Yet, Ohno et al. reported contradictory findings in patients in whom higher ADC on pretreatment MRI were significantly associated with poor prognosis [38]. The discrepancies in the predictive power of the ADC may in part be due to the non-uniform measurement. Depending on the study, the mean, minimum or maximum ADC value is used. Furthermore, until now there is no clear definition of where within the tumor one should place the ADC ROI. Further studies are needed to establish a uniform and reproduceable measurement of the ADC and thus to substantiate its prognostic value.
Beside of all of these capabilities MRI offers in imaging of the NSCLC, the acquisition time of this modality has to be viewed critically, especially in comparison with CT. As mentioned in the first section of the discussion, the MRI protocol takes about 15–30 minutes. Compared to a CT scan of the thorax with an acquisition time of only a few seconds, this is of course a considerable effort, especially for patients with impaired lung function. However, the MRI protocol can certainly be optimized by removing, respectively limiting the time-consuming breath-triggered T2-TSE to the target areas, because acquisition of the whole thorax can take between 15–30 minutes depending on the patient's body height and breathing variability. If assessment of the ultra-structure of the lung is not required, DWI/ADC in combination with T2-HASTE, both only taking about 5 min for image acquisition, could be a solution for thorax imaging regarding T and N stadium.
Our study had some limitations. First, it is a single center study with a small number of patients. Additionally, some patients did not complete scanning schedule and we can’t exclude the possibility that this might have skewed the results.