MRI examinations were performed on 62 patients with GIST on 1.5T scanners before and at two and 12 weeks after treatment with targeted agents. The longest diameter (LD) and contrast-to-noise ratio (CNR) of the tumors were measured with T2-weighted imaging (T2WI), and the apparent diffusion coefficient (ADC) was measured with diffusion-weighted imaging (DWI, b = 0 and 1000 s/mm²).

Patients with a maximum of two lesions per organ/site and five lesions total were enrolled, and the quantitative parameters were averaged according to RECIST 1.1 criteria. The patients were classified as having a good response (GoodR) if the target lesions exhibited a 10% or greater reduction in the longest diameter or displayed apparent cystic or myxoid degeneration after three months of therapy; otherwise, they were considered as having a poor response (PoorR), according to the combined Choi and EORTC-ISG-AGITG criteria ^[1].

The pre-therapy values and early percentage changes (%Δ) of the three averaged parameters were compared regarding their ability to differentiate the good response (GoodR) from the poor response (PoorR) cases, using receiver operating characteristic (ROC) curves.

A total of 62 patients with 141 lesions were enrolled in the study. There were 42 patients in the GoodR group and 20 in the PoorR group.

No significant differences in the three baseline parameters were found between the GoodR and PoorR groups (ADC: 1.18 ± 0.27× 10^-3 mm²/s vs. 1.17 ± 0.31× 10^-3 mm²/s, P = 0.946; LD: 70.58 ± 33.40 mm vs. 59.16 ± 40.55 mm, P = 0.245; T2WI-CNR: 39.28 ± 22.62 vs. 30.32 ± 18.19, P = 0.127). After two weeks of therapy, the percentage changes in the ADC and LD were significantly different between the two groups (ADC: GoodR 30% vs. PoorR 1%, Z = -4.819, P < 0.001; LD: GoodR -7% vs. PoorR -2%, Z = -3.238, P = 0.001), but percentage changes in T2WI-CNR were not significantly different (GoodR -3% vs. PoorR 9%, Z = -0.663, P = 0.508). ROC curves revealed that the AUCs for the pre-therapy LD, T2WI-CNR and ADC were 0.644, 0.615 and 0.508, respectively (Figure 1), whereas the AUCs for the percentage changes of LD, T2WI-CNR and ADC after two weeks of therapy were 0.756, 0.552 and 0.881, respectively (Figure 2), in the differentiation of response groups. When %ΔADC ≥ 15% was taken to predict a GoodR, the PPV was 93.3% (a typical case is shown in Figure 3); when %ΔADC ≤ 1% was taken to predict a PoorR, the PPV was 85.7% (a typical case is shown in Figure 4).

Recent studies have confirmed the potential value of DWI predicting the responses of malignant tumors to anticancer therapies ^[2]. To our knowledge, this study is the first patient-based research investigating the response prediction of GIST to targeted therapy using DWI, which has better clinical applicability than the previous lesion-based studies.

We investigated the practical threshold of %ΔADC to provide additional information for clinical decision making. An increase in %ΔADC ≥ 15% after two weeks of therapy indicated a likely good response, with a PPV of 93.3%; this response suggests that the patients could continue their initial treatment protocol. On the contrary, if the ADC after two weeks of therapy showed no explicit change (%ΔADC ≤ 1%), then a poor response was highly suspected; however, considering there was still a 15% false-positive rate, a shortening of the follow-up time intervals is suggested to further confirm these findings.

In our research, no significant difference was found between the GoodR and the PoorR groups regarding pre-therapy ADC, which was different from the previous lesion-based imatinib studies ^[3]. We postulate that the mixed patient cohort examined in our study might partly explain this result. Additionally, the patients who accepted Sutent treatment usually experienced progression in the regions that were previously necrotic or cystic, which exhibited a heterogeneous signal background and led to an increase in the pre-therapy ADC. However, the performance of the percentage change in the ADC after two weeks of therapy was not affected and it was much better than the baseline ADC. Therefore, it is worthwhile to administer a short-term targeted agent as an experimental treatment to obtain more accurate response prediction results.