Tumor habitat characteristics derived from intravoxel incoherent motion for early response assessment in soft tissue sarcoma undergoing neoadjuvant radiotherapy and targeted therapy: a phase II study

Introduction

Soft tissue sarcomas (STSs) are a diverse group of malignant tumors originating from mesenchymal tissues with over 100 subtypes and constituting less than 1% of adult solid tumors (1-4). Neoadjuvant radiotherapy (neo-RT) is increasingly preferred for its smaller target volumes, reduced radiation doses, and diminished risks of late toxicity (1,5). One type of targeted drug, tyrosine kinase inhibitors, can be effective as radio-sensitizing agents. They work by blocking tumor-related pathways, which in turn normalizes the vasculature and improves tumor oxygenation (6,7). Studies have proven that the combination of radiotherapy (RT) and targeted therapy can increase pathological complete response (pCR) rates (8-10). However, neo-RT may increase surgical complexity and postoperative complication risks (5,11). Early treatment evaluation is crucial for personalized therapy, but current criteria like Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 (12) often underperform in STS due to post-treatment intratumoral necrosis, hemorrhage, and cystic changes (13,14), and pathological response degree remains the most objective, reliable indicator (15). To address the shortcomings of current evaluation methods, a precise, non-invasive method to accurately assess pathological response is urgently needed.

Magnetic resonance imaging (MRI), with its high resolution and diverse sequences, is a leading diagnostic tool for STS localization and qualitative assessments (1). Diffusion-weighted imaging (DWI) has demonstrated the ability to predict treatment effects in STS patients (16,17). However, conventional DWI relies on a mono-exponential model assuming unrestricted water diffusion, which fails to capture non-Gaussian water motion in biological tissues (18). In contrast, intravoxel incoherent motion (IVIM), an advanced diffusion model, provides quantitative information on microvascular perfusion and water diffusion within tissues (19). IVIM has been used for diagnosis, treatment evaluation, and recurrence detection in various cancers (20), including breast (21) and rectal (22) cancers.

Tumor heterogeneity, driven by internal microenvironment differences, is critical for staging and prognosis (23). Previous research has predominantly concentrated on whole-tumor radiomics, neglecting the intra-tumor ecological diversity (24,25). To address this limitation, tumor habitat analysis has become a potential method (26). It groups similar voxels into distinct subregions based on shared biological characteristics (27), quantifying internal tumor heterogeneity and reflecting local microenvironment differences (28,29). Subregional tumor habitats have shown utility in predicting treatment responses for various tumors (23,30-32). However, their application in STS remains unreported.

Therefore, this study aimed to use tumor subregional IVIM parameters to assess pathological response degree in STS patients undergoing neo-RT and targeted therapy. We present this article in accordance with the STARD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2201/rc).

Methods

Participants

This study was approved by the Ethics Committee at the Cancer Hospital, Chinese Academy of Medical Sciences (No. NCC-4903), and written informed consent was obtained from all patients. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Patients were recruited from a prospective phase II trial (*NCT05167994/ChiCTR2000033377 and NCT05235100) at the Department of Radiotherapy, Cancer Hospital, Chinese Academy of Medical Sciences, from October 2020 to September 2024, with a series of prospective MRI images acquired both prior to and subsequent to neo-RT and targeted therapy. The specific inclusion and exclusion criteria, along with the treatment and evaluation of clinical efficacy, are detailed in Appendix 1. Figure 1 demonstrates the flow diagram of the study cohort.

Figure 1 Flow of study cohort inclusion. MRI, magnetic resonance imaging; STS, soft tissue sarcoma.

MRI image acquisition

All patients received MRI scans before and one month after neo-RT with targeted therapy utilizing a 3-T scanner (SIGNA Premier; GE Healthcare, Waukesha, WI, USA). Depending on the anatomical region, we used either a 30-channel phased-array flexible coil or a 16-channel extremity coil. The scanning protocol, at minimum, included sequences of nonenhanced T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and IVIM. Enhanced T1WI was included based on the individual clinical requirements. Details of our IVIM sequences are provided in Appendix 1.

Tumor segmentation and IVIM image processing

A radiologist (with 4 years of STS experience in image diagnosis) manually outlined the region of interest (ROI) slice-by-slice on the IVIM with b=800 along the edge of the whole tumor for each patient by using ITK-SNAP (version 4.2.0; www.itksnap.org). All segmentations were checked by a senior radiologist with 23 years of STS experience, blinded to clinical and pathological data. ROIs encompassed the entire tumor, excluding peritumoral edema and signal outliers, to minimize partial-volume effects. IVIM parameter maps, including true-diffusion coefficient (D), pseudo-diffusion coefficient (D*), and perfusion fraction (f), were calculated by a custom MATLAB script (R2021b, MathWorks, Natick, MA).

Spatial tumor habitats and IVIM parameter extraction

Tumor habitat clustering was performed using proprietary Python-based software (version 3.9.12; scikit-learn package). IVIM-D and IVIM-D* maps were selected as data inputs for clusters, as they effectively represent water molecule diffusion and microcirculatory perfusion within tissues, respectively (19). Since both D and D* maps originate from the same DWI data, no additional image registration was required. All parameter maps were resampled to isotropic voxel sizes of 3×3×3 mm³ before analysis to ensure uniformity. K-means clustering was used to divide the entire tumor into subregional habitats, with the optimal number of clusters determined by silhouette score (Figure S1) (33). Details of cluster selection are provided in Appendix 1.

Mean IVIM parameter values for the whole tumor and subregions were extracted pre- and post-treatment using PyRadiomics (version 3.1.0) on the Python platform (version 3.7.9; https://www.python.org), offering quantitative insights into tumor heterogeneity and treatment response.

Model construction

Postoperative pathological findings classified patients into good-response (<50% residual tumor cells) and poor-response (≥50% residual tumor cells) groups (13). This cutoff was also determined based on preliminary unpublished data showing its correlation with distant metastasis-free survival (Figure S2).

Univariate analyses (t-tests or Mann-Whitney U tests) were performed to identify features significantly associated with the response groups (P<0.05). These features were then used to establish univariate logistic regression models. Subsequently, a combined model was developed by applying least absolute shrinkage and selection operator logistic regression with L1 regularization. Treatment efficacy was also assessed in accordance with the RECIST 1.1 criteria by using T1WI and T2WI (12). Figure 2 demonstrated the workflow of this study.

Figure 2 The workflow of this study. D, true-diffusion coefficient; D*, pseudo-diffusion coefficient; ROI, region of interest.

Statistical analysis

Statistical analyses were performed utilizing R (version 4.4.1; http://www.rproject.org). The Shapiro-Wilk test was employed to assess data normality. Categorical variables were compared using Chi-squared or Fisher’s exact test, while continuous variables were analyzed with the t-test and Mann-Whitney U test. Differences in IVIM parameters extracted from whole tumor and subregions between good- and poor-response groups were assessed using t-test and Mann-Whitney U test according to normality. Receiver operating characteristic (ROC) curve analysis was performed to assess diagnostic performance, with optimal cut-off point determined using the Youden index. Performance metrics, including accuracy, sensitivity, and specificity, were calculated. The DeLong test was utilized to compare the area under curve (AUC) values of ROC curves among all models. Meanwhile, the goodness-of-fit of the model was evaluated using the Hosmer-Lemeshow test, while model stability and reliability were verified using calibration curves generated through Bootstrap resampling (n=1,000). Moreover, 200 iterations of 5-fold cross-validation were performed to calculate and correct the biased AUC. Decision curve analysis was employed to assess the clinical utility of predictive models across different threshold probabilities, assessing net benefits relative to “treat-all” and “treat-none” strategies.

Results

Patient characteristics

The flow of this study is depicted in Figure 1. Table 1 summarizes the inclusion of 21 patients (11 males and 10 females; mean age, 52.1±17.4 years). According to their postoperative pathologies, 13 (61.9%) patients demonstrated good treatment response (Figure 2), whereas the remaining 8 (38.1%) patients had poor treatment response (Figure 3). Neo-RT was combined with apatinib in 9 patients and with anlotinib in 12 patients.

Figure 3 A 68-year-old woman with STS exhibiting good responses to neo-RT and targeted therapy. Images in sequences are pre-treatment and post-treatment MRI imaging and IVIM parametric maps. Pre-treatment DWI (A) and post-treatment DWI (E) showed mixed high and low signals, predominantly high signals STS at its largest dimension. Corresponding pre-treatment D (B), pre-treatment D* (C), pre-treatment f (D), post-treatment D (F), post-treatment D* (G), and post-treatment f (H) maps are shown. The same color for a parameter before and after therapy indicates the same value. D, true diffusion coefficient; D*, pseudo-diffusion coefficient; DWI, diffusion-weighted imaging; f, perfusion fraction; IVIM, intravoxel incoherent motion; MRI, magnetic resonance imaging; neo-RT, neoadjuvant radiotherapy; STS, soft tissue sarcoma.

Feature selection

Feature selection in whole tumors

We compared the mean D, D*, and f values of whole tumors in both pre- and post-neo-RT and targeted therapy imaging. Notably, D of post- neo-RT and targeted therapy, denoted as D_{whole Mean (Post)}, significantly differed between the good- and poor-response groups (P<0.05). Moreover, D_{whole Mean (Post)} was higher in the good-response group compared to the poor-response group (Table 2).

Feature selection in three subregional tumor habitat clusters

The optimal cluster number is 3 based on the silhouette score. The levels of voxel distribution in the IVIM-D and IVIM-D* maps (K=3) are shown in Figure 4. Three subregions were defined based on D and D* (Figure 5): Cluster 1, a region with lower D* and higher D (indicating lower blood perfusion or microvascular density with lower cellular density or larger cell spacing, where water molecules diffuse freely); Cluster 2, a region with higher D* (signifying higher blood perfusion or microvascular density); and Cluster 3, a region with lower D* and D (suggesting lower blood perfusion or microvascular density but higher cellular density or narrower cell spacing, where water molecule diffusion is restricted).

Figure 4 A 36-year-old woman with STS exhibited a poor response to neo-RT and targeted therapy. Images in sequence are pre-treatment and post-treatment MRI and IVIM parametric maps. After the treatment, the maximum diameter of the tumor decreased, but the pathological result showed a tumor residual rate of 80% (>50%). Pre-treatment (A) and post-treatment (E) DWI demonstrated mixed high and low signals. Corresponding pre-treatment D map (B), pre-treatment D* (C), pre-treatment f (D), post-treatment D (F), post-treatment D* (G), and post-treatment f (H) maps are shown. The same color for one parameter before and after therapy indicates the same value. D, true diffusion coefficient; D*, pseudo-diffusion coefficient; DWI, diffusion-weighted imaging; f, perfusion fraction; IVIM, intravoxel incoherent motion; MRI, magnetic resonance imaging; neo-RT, neoadjuvant radiotherapy; STS, soft tissue sarcoma.

Figure 5 K-means clustering applied to D and D* images to construct MRI habitats. (A) Scatter plot illustrating the distribution of parameters D and D* for clustered voxels. Blue, orange, and green areas represent Cluster 1, Cluster 2 and Cluster 3, respectively. (B) Three subregions were defined based on D and D*: Cluster 1, a region with lower D* and higher D; Cluster 2, a region with higher D*; Cluster 3, a region with lower D* and lower D. D, true diffusion coefficient; D*, pseudo-diffusion coefficient; MRI, magnetic resonance imaging.

Among the mean values of D, D*, and f in three subregional tumor habitats before and after neo-RT, four features exhibited significant differences between the good-response and poor-response groups (P<0.05): D_{Cluster 1 Mean (Post)}, D*_{Cluster 2 Mean (Pre)}, f_{Cluster 1 Mean (Post)}, and f_{Cluster 3 Mean (Post)}, and the four parameters are higher in the good-response group than in the poor-response group (Table 2). Among them, only D_(Post) [i.e., D_{whole Mean (Post)} and D_{Cluster 1 Mean (Post)}] was significant in both the whole tumor and its subregions. D*_{whole Mean (Pre)} and f_{whole Mean (Post)} were nonsignificant in the whole tumor; nevertheless, they were significant within the subregions after clustering [D*_{Cluster 2 Mean (Pre)}, f_{Cluster 1 Mean (Post)}, and f_{Cluster 3 Mean (Post)}].

Model construction and evaluation

Five selected [D_{whole Mean (Post)}, D_{Cluster 1 Mean (Post)}, D*_{Cluster 2 Mean (Pre)}, f_{Cluster 1 Mean (Post)} and f_{Cluster 3 Mean (Post)}] parameters were used to construct individual parameter models, which all outperformed RECIST 1.1 (AUC =0.567). Specifically, the AUC values for the individual parameter models were as follows: 0.808 of D_{whole Mean (Post)}, 0.837 for D_{Cluster 1 Mean (Post)}, 0.856 for D*_{Cluster 2 Mean (Pre)}, 0.865 for f_{Cluster 1 Mean (Post)}, and 0.779 for f_{Cluster 3 Mean (Post)}. Notably, the combined model demonstrated the highest assessment performance, with an AUC of 0.904, significantly outperforming all other individual models. A detailed comparison of the AUC of all models, evaluated using the DeLong method, is provided in Table S1. The detailed performance metrics of each model are summarized in Table 3, and the corresponding ROC curves are summarized in Figure 6. Calibration curves for all models are presented in Figures S3-S9. Decision curve analysis indicated that the D*_{Cluster 2 Mean (Pre)} model provided the highest net benefit at low-risk thresholds (<0.3), whereas the combined model demonstrated the highest net benefit at medium-risk thresholds (0.4–0.6). Finally, most models exhibited significant fluctuations in the net benefit at high-risk thresholds (>0.7; Figure S10).

Figure 6 Comparison of the predictive performances of all models. AUC, area under curve; D, true diffusion coefficient; D*, pseudo-diffusion coefficient; f, perfusion fraction; RECIST, Response Evaluation Criteria in Solid Tumors.

Discussion

This study evaluated the applicability of IVIM-derived tumor habitat parameters (D, D*, and f) for early assessment of treatment efficacy in STS patients undergoing neo-RT and targeted therapy. The results demonstrated that D_{whole Mean (Post)}, D_{Cluster 1 Mean (Post)}, D*_{Cluster 2 Mean (Pre)}, f_{Cluster 1 Mean (Post)}, and f_{Cluster 3 Mean (Post)} significantly differed between the good- and poor-response groups. All IVIM individual parameter models outperformed RECIST 1.1 and the combined model had better performance than RECIST 1.1 and all other individual parameter models. This superiority is likely because RECIST 1.1 relies solely on unidimensional tumor size changes, whereas IVIM-derived parameters capture early microstructural and microvascular alterations and intra-tumoral heterogeneity that may precede apparent tumor shrinkage. From a clinical perspective, an IVIM-based composite model that surpasses RECIST 1.1 could enable earlier and more accurate identification of good and poor responders, thereby supporting individualized treatment adaptation, such as timely modification of systemic therapy and optimization of surgical planning in STS patients undergoing neo-RT with targeted therapy.

Compared with conventional imaging methods such as T2WI and/or T1WI, dynamic contrast-enhanced MRI (24,25,34) and mono-exponential DWI model (16), we applied IVIM parameters to assess the effectiveness of neo-RT and targeted therapy for STS, offering a more precise acquisition of the non-Gaussian diffusion features of water molecules within tumor tissues (19). Furthermore, previous studies were mainly based on the analysis of the entire tumor, neglecting tumoral heterogeneity (25). Alternatively, newly-emerging habitat imaging methods have been utilized to explore the intra-tumoral heterogeneity, thus addressing the limitation of previous research (35). Benvenuti et al. found that histogram and clustering analysis based on MRI can provide potential imaging indicators for identifying between patients with high-grade osteosarcoma who have a poor response and those with a fair-to-good response to neoadjuvant chemotherapy (31). Similarly, our analysis employed IVIM parameters D and D* to delineate physiological subregions of tumors using a data-driven clustering method, rather than treating tumors as homogeneous masses (24,25). This approach reflects the tumor’s spatial heterogeneity and reveals distinct physiological properties, such as cellular density, vascularization, and perfusion.

In our study, D_{whole Mean (Post)}, D_{Cluster 1 Mean (Post)}, D*_{Cluster 2 Mean (Pre)}, f_{Cluster 1 Mean (Post)} and f_{Cluster 3 Mean (Post)} in the good-response group were significantly higher compared to those in the poor-response group. Wan et al. demonstrated that D outperformed other IVIM parameters in tracking anti-vascular therapy responses, which provides a relevant context for our current findings (30). Treatment-induced decreases in tumor cells (8) lead to increased water molecule diffusion. In comparison to the poor-response group, the good-response one demonstrated lower tumor residential rates and tumor cell densities; consequently, there was a more obvious increase in the degree of diffusion, as reflected by higher D_{whole Mean (Post)} and D_{Cluster 1 Mean (Post)} in the good-response group than in the poor-response group. Additionally, regions of tumors characterized by low perfusion exhibit an increased likelihood of hypoxia or necrosis. Furthermore, the disintegration of membrane structures can enhance diffusion through tissue degradation (36). Cluster 1, characterized by lower D* and higher D, represents areas with reduced blood perfusion, lower microvascular density, and low cellular density, resulting in increased water diffusion. This cluster may correspond to regions more susceptible to treatment effects, including tumor cell reduction and enhanced water molecule mobility.

The parameter f, representing the proportion of microcirculatory perfusion within the overall diffusion, can indicate tumor vascular normalization (18,30). Vascular endothelial growth factor receptor 2 blockade can normalize tumor vessel structure temporarily, thereby enhancing vascular functionality (i.e., tumor oxygenation) and RT response (37,38). The higher f values observed in the good-response group suggest a higher degree of vascular normalization induced by effective treatment. Tumor regions with lower perfusion are particularly susceptible to hypoxia or necrosis (36), potentially corresponding to Clusters 1 and 3 with decreased blood supply (lower D*) and characterized by abnormal vasculature, and the normalization of vessels in these areas may be affected by treatment substantially.

Cancer treatment can reduce blood supply to a tumor, along with blood perfusion and micro-vessel density (8), thereby limiting tumor growth and spread. D* indicates the diffusion movement associated with microcirculation perfusion within a voxel, and Cluster 2 is characterized by higher D*, indicating increased blood perfusion or microvascular density. Cell proliferation requires adequate perfusion (39). Thus, Cluster 2 may correspond spatially to the histological regions demonstrating increased proliferative activity (36), and this cluster may be more sensitive to changes in blood supply due to treatment, there were significant disparities in the D*_{Cluster 2 Mean (Pre)} values between the groups with good response and those with poor response.

This study carefully integrated ecological diversity characteristics within tumors with imaging features, thus enabling a far more comprehensive and accurate assessment than imaging features alone. Of all IVIM parameters (D, D*, and f) in the whole tumors and their subregions, only D_(Post) [i.e., D_{whole Mean (Post)} and D_{Cluster 1 Mean (Post)}] was significant in both whole tumors and subregions. Differences in D*_(Pre) and f_(Post) were nonsignificant in the whole tumors but significant within the subgroups after clustering [D*_{Cluster 2 Mean (Pre)}, f_{Cluster 1 Mean (Post)}, and f_{Cluster 3 Mean (Post)}]. What’s more, our results demonstrated that the AUC of the D_{whole Mean (Post)} model for neo-RT and targeted therapy response assessment was lower than that of D_{Cluster 1 Mean (Post)} (0.808 vs. 0.837; P < 0.05). This significant finding validates that the diagnostic model, which is established with features carefully extracted from different habitats, demonstrated notably higher precision and enhanced diagnostic performance.

Recent studies have highlighted the programmed cell death protein 1 (PD-1) and PD-1 ligand-1 (PD-L1) immune checkpoint as an emerging therapeutic target in bone sarcomas and STSs, and immune checkpoint inhibitors (ICIs) have shown limited but encouraging activity in selected subtypes. In parallel, preclinical and early clinical data suggest that RT and targeted agents can enhance antitumor immunity by promoting immunogenic cell death, antigen release, and modulation of immune-related signaling pathways (40,41). Taken together, these mechanisms provide a biological rationale for exploring the integration of PD-1/PD-L1 blockade with neo-RT plus targeted therapy in high-grade STS, with the aim of amplifying systemic immune responses and potentially overcoming primary resistance to ICIs (42). On this basis, our neo-RT plus targeted therapy regimen could be integrated with ICIs in future trials, and IVIM-derived tumor habitats may provide non-invasive imaging biomarkers to select patients and monitor early response to such multimodal, immunotherapy-based treatments.

This study also has several limitations. First, this research was limited to a single center and the sample size of patients was comparatively small. Thus, subsequent research focused on refining or validating our models is warranted. Second, the selection of b-values plays an essential part in influencing the fitting results of IVIM model parameters. However, so far, there are still no universally accepted criteria for determining the optimal range of b-values. Current choices in treating STS largely rely on empirical experience. Thus, it’s crucial to systematically explore and precisely determine the most suitable b-value range for STS. Moreover, although one radiologist delineated the ROIs and another senior radiologist examined them, the process was manual and, therefore, may have led to partial bias.

Conclusions

IVIM-derived parameters and tumor habitat analysis offer valuable insights as non-invasive imaging biomarkers for assessing treatment response to neo-RT and targeted therapy in STS capturing the spatial heterogeneity and distinct microenvironmental features within tumors.

Acknowledgments

We would like to thank the investigators at all participating study sites.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2201/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2201/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2201/prf

Funding: This work was supported by the CAMS Innovation Fund for Medical Sciences (CIFMS) (grant Nos. 2023-I2M-C&T-B-089; 2025-I2M-C&T-B-055), the Beijing Hope Run Special Fund of Cancer Foundation of China and National High Level Hospital Clinical Research Funding (Nos. 2022-CICAMS-80102022203; LC2020A15; LC2024A11), and the National High Level Hospital Clinical Research Funding and National Cancer Center Climbing Fund (NCC202416001).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2201/coif). S.W. is an employee of GE Healthcare and reports data analysis support from GE Healthcare. Y.J. and Y.S. are an employee of Siemens Healthineers; they report that Siemens Healthineers Ltd. is an MR collaboration scientist doing technical support in this study under Siemens collaboration regulation; all the corporations involved had no influence on the design, conduct, or interpretation of the results, nor financial support of this study. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The Ethics Committee at the Cancer Hospital, Chinese Academy of Medical Sciences approved this prospective study (No. NCC-4903), and informed consent was taken from all individual participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Casali PG, Abecassis N, Aro HT, et al. Soft tissue and visceral sarcomas: ESMO-EURACAN Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2018;29:iv51-67. [Crossref] [PubMed]
2. Wibmer C, Leithner A, Zielonke N, et al. Increasing incidence rates of soft tissue sarcomas? A population-based epidemiologic study and literature review. Ann Oncol 2010;21:1106-11. [Crossref] [PubMed]
3. Sbaraglia M, Bellan E, Dei Tos AP. The 2020 WHO Classification of Soft Tissue Tumours: news and perspectives. Pathologica 2021;113:70-84. [Crossref] [PubMed]
4. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49. [Crossref] [PubMed]
5. Noeuveglise M, Tessier W, Barthoulot M, et al. Preoperative versus postoperative radiotherapy for localized soft tissue sarcoma treated with curative intent in a French tertiary center "SARCLOC". BMC Cancer 2024;24:1550. [Crossref] [PubMed]
6. Yuan J, Li X, Yu S. Molecular targeted therapy for advanced or metastatic soft tissue sarcoma. Cancer Control 2021;28:10732748211038424. [Crossref] [PubMed]
7. George S, Merriam P, Maki RG, et al. Multicenter phase II trial of sunitinib in the treatment of nongastrointestinal stromal tumor sarcomas. J Clin Oncol 2009;27:3154-60. [Crossref] [PubMed]
8. Yoon SS, Duda DG, Karl DL, et al. Phase II study of neoadjuvant bevacizumab and radiotherapy for resectable soft tissue sarcomas. Int J Radiat Oncol Biol Phys 2011;81:1081-90. [Crossref] [PubMed]
9. Canter RJ, Borys D, Olusanya A, et al. Phase I trial of neoadjuvant conformal radiotherapy plus sorafenib for patients with locally advanced soft tissue sarcoma of the extremity. Ann Surg Oncol 2014;21:1616-23. [Crossref] [PubMed]
10. Lewin J, Khamly KK, Young RJ, et al. A phase Ib/II translational study of sunitinib with neoadjuvant radiotherapy in soft-tissue sarcoma. Br J Cancer 2014;111:2254-61. [Crossref] [PubMed]
11. Gamboa AC, Gronchi A, Cardona K. Soft-tissue sarcoma in adults: An update on the current state of histiotype-specific management in an era of personalized medicine. CA Cancer J Clin 2020;70:200-29. [Crossref] [PubMed]
12. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45:228-47. [Crossref] [PubMed]
13. Stacchiotti S, Collini P, Messina A, et al. High-grade soft-tissue sarcomas: tumor response assessment--pilot study to assess the correlation between radiologic and pathologic response by using RECIST and Choi criteria. Radiology 2009;251:447-56. [Crossref] [PubMed]
14. Spinnato P, Kind M, Le Loarer F, et al. Soft Tissue Sarcomas: The Role of Quantitative MRI in Treatment Response Evaluation. Acad Radiol 2022;29:1065-84. [Crossref] [PubMed]
15. Wang D, Harris J, Kraybill WG, et al. Pathologic Complete Response and Clinical Outcomes in Patients With Localized Soft Tissue Sarcoma Treated With Neoadjuvant Chemoradiotherapy or Radiotherapy: The NRG/RTOG 9514 and 0630 Nonrandomized Clinical Trials. JAMA Oncol 2023;9:646-55. [Crossref] [PubMed]
16. Gao Y, Ghodrati V, Kalbasi A, et al. Prediction of soft tissue sarcoma response to radiotherapy using longitudinal diffusion MRI and a deep neural network with generative adversarial network-based data augmentation. Med Phys 2021;48:3262-372. [Crossref] [PubMed]
17. Gao Y, Kalbasi A, Hsu W, et al. Treatment effect prediction for sarcoma patients treated with preoperative radiotherapy using radiomics features from longitudinal diffusion-weighted MRIs. Phys Med Biol 2020;65:175006. [Crossref] [PubMed]
18. Le Bihan D, Breton E, Lallemand D, et al. MR imaging of intravoxel incoherent motions: application to diffusion and perfusion in neurologic disorders. Radiology 1986;161:401-7. [Crossref] [PubMed]
19. Le Bihan D, Breton E, Lallemand D, et al. Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging. Radiology 1988;168:497-505. [Crossref] [PubMed]
20. Iima M. Perfusion-driven Intravoxel Incoherent Motion (IVIM) MRI in Oncology: Applications, Challenges, and Future Trends. Magn Reson Med Sci 2021;20:125-38. [Crossref] [PubMed]
21. Zhang J, Zheng Y, Li L, et al. Combination of IVIM with DCE-MRI for diagnostic and prognostic evaluation of breast cancer. Magn Reson Imaging 2024;113:110204. [Crossref] [PubMed]
22. Zhang L, Jin Z, Yang F, et al. Added value of histogram analysis of intravoxel incoherent motion and diffusion kurtosis imaging for the evaluation of complete response to neoadjuvant therapy in locally advanced rectal cancer. Eur Radiol 2025;35:1669-78. [Crossref] [PubMed]
23. O'Connor JP, Rose CJ, Waterton JC, et al. Imaging intratumor heterogeneity: role in therapy response, resistance, and clinical outcome. Clin Cancer Res 2015;21:249-57. [Crossref] [PubMed]
24. Peeken JC, Asadpour R, Specht K, et al. MRI-based delta-radiomics predicts pathologic complete response in high-grade soft-tissue sarcoma patients treated with neoadjuvant therapy. Radiother Oncol 2021;164:73-82. [Crossref] [PubMed]
25. Miao L, Cao Y, Zuo L, et al. Predicting pathological complete response of neoadjuvant radiotherapy and targeted therapy for soft tissue sarcoma by whole-tumor texture analysis of multisequence MRI imaging. Eur Radiol 2023;33:3984-94. [Crossref] [PubMed]
26. Wu Y, Zhang W, Liang X, et al. Habitat radiomics analysis for progression free survival and immune-related adverse reaction prediction in non-small cell lung cancer treated by immunotherapy. J Transl Med 2025;23:393. [Crossref] [PubMed]
27. Torheim T, Groendahl AR, Andersen EK, et al. Cluster analysis of dynamic contrast enhanced MRI reveals tumor subregions related to locoregional relapse for cervical cancer patients. Acta Oncol 2016;55:1294-8. [Crossref] [PubMed]
28. Zhang Y, Chen J, Yang C, et al. Preoperative prediction of microvascular invasion in hepatocellular carcinoma using diffusion-weighted imaging-based habitat imaging. Eur Radiol 2024;34:3215-25. [Crossref] [PubMed]
29. Sala E, Mema E, Himoto Y, et al. Unravelling tumour heterogeneity using next-generation imaging: radiomics, radiogenomics, and habitat imaging. Clin Radiol 2017;72:3-10. [Crossref] [PubMed]
30. Wan Q, Bao Y, Xia X, et al. Intravoxel Incoherent Motion Diffusion-Weighted Imaging for Predicting and Monitoring the Response of Anti-Angiogenic Treatment in the Orthotopic Nude Mouse Model of Lung Adenocarcinoma. J Magn Reson Imaging 2022;55:1202-10. [Crossref] [PubMed]
31. Benvenuti G, Marzi S, Vidiri A, et al. Prediction of tumor response to neoadjuvant chemotherapy in high-grade osteosarcoma using clustering-based analysis of magnetic resonance imaging: an exploratory study. Radiol Med 2025;130:13-24. [Crossref] [PubMed]
32. Huang YH, Shi ZY, Zhu T, et al. Longitudinal MRI-Driven Multi-Modality Approach for Predicting Pathological Complete Response and B Cell Infiltration in Breast Cancer. Adv Sci (Weinh) 2025;12:e2413702. [Crossref] [PubMed]
33. Zhang W, Wu L, Zhang S. Clinical phenotype of ARDS based on K-means cluster analysis: A study from the eICU database. Heliyon 2024;10:e39198. [Crossref] [PubMed]
34. Crombé A, Périer C, Kind M, et al. T(2) -based MRI Delta-radiomics improve response prediction in soft-tissue sarcomas treated by neoadjuvant chemotherapy. J Magn Reson Imaging 2019;50:497-510. [Crossref] [PubMed]
35. Li S, Dai Y, Chen J, et al. MRI-based habitat imaging in cancer treatment: current technology, applications, and challenges. Cancer Imaging 2024;24:107. [Crossref] [PubMed]
36. Jalnefjord O, Montelius M, Arvidsson J, et al. Data-driven identification of tumor subregions based on intravoxel incoherent motion reveals association with proliferative activity. Magn Reson Med 2019;82:1480-90. [Crossref] [PubMed]
37. Winkler F, Kozin SV, Tong RT, et al. Kinetics of vascular normalization by VEGFR2 blockade governs brain tumor response to radiation: role of oxygenation, angiopoietin-1, and matrix metalloproteinases. Cancer Cell 2004;6:553-63. [Crossref] [PubMed]
38. Gasparini G, Torino F, Ueno T, et al. A phase II study of neoadjuvant bevacizumab plus capecitabine and concomitant radiotherapy in patients with locally advanced rectal cancer. Angiogenesis 2012;15:141-50. [Crossref] [PubMed]
39. De Luise M, Kurelac I, Coluccelli S, et al. Perfusion-based ex vivo culture of frozen ovarian cancer tissues with preserved tumor microenvironment. NPJ Precis Oncol 2025;9:152. [Crossref] [PubMed]
40. Fazel M, Dufresne A, Vanacker H, et al. Immunotherapy for Soft Tissue Sarcomas: Anti-PD1/PDL1 and Beyond. Cancers (Basel) 2023;15:1643. [Crossref] [PubMed]
41. Hashimoto K, Nishimura S, Goto K. PD-1/PD-L1 immune checkpoint in bone and soft tissue tumors Mol Clin Oncol 2025;22:31. (Review). [Crossref] [PubMed]
42. Procureur A, Simonaggio A, Bibault JE, et al. Enhance the Immune Checkpoint Inhibitors Efficacy with Radiotherapy Induced Immunogenic Cell Death: A Comprehensive Review and Latest Developments. Cancers (Basel) 2021;13:678. [Crossref] [PubMed]