Multi-b-value diffusion-weighted imaging-based habitat imaging for preoperative prediction of histological grade and prognosis in soft tissue sarcoma: focusing on intratumoral heterogeneity Authors: Chunlian Chen, Jie Zhou, Zesheng Zhang, Kai Zhang, Xu Wang, Lei Shi, Yuxin Pan, Kexin Zhang, Huimin Ye, Zi-Ang Pan, Yixin Liu, Wenjia Wang, Shaowu Wang Published: 2026 ABSTRACT AbstractBackgroundMarked intratumoral heterogeneity (ITH) is a critical biological characteristic of soft tissue sarcoma (STS). This study aimed to develop a novel multi-b-value diffusion-weighted imaging (DWI)-based habitat imaging approach to quantify and visualize ITH for the preoperative prediction of the histological grade and prognosis of STS.MethodsA total of 101 patients were prospectively enrolled in this study and underwent multi-b-value DWI examinations. The true diffusion coefficient (D), perfusion fraction (f), and mean kurtosis coefficient (MK) parameter maps were clustered using the K-means algorithm. The elbow method and silhouette coefficient were employed to determine the appropriate number of habitats. A modified histological grading system incorporating Ki‑67 expression was adopted to stratify the patients into low‑ and high‑grade groups. Differences in clinical features, conventional magnetic resonance imaging (MRI) features, and habitat features between the two groups were compared using the Chi‑squared test, independent‑samples t‑test, or Mann-Whitney U test, as appropriate. Predictive models were developed using univariate and multivariable logistic regression. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, and model calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test. Cox proportional hazards regression and Kaplan-Meier analyses were performed to identify risk factors for overall survival (OS) and metastasis‑free survival (MFS).ResultsFour habitats were identified. Several features derived from Habitats 1 and 3 differed significantly between the low- and high-grade groups (P<0.05). Based on the grading predictors identified by logistic regression, both the combined model [area under the curve (AUC) =0.896, accuracy =87.28%, sensitivity =79.70%, specificity =80.95%, 95% confidence interval (CI): 0.837–0.916] and habitat model (AUC =0.841, accuracy =76.75%, sensitivity =69.50%, specificity =85.71%, 95% CI: 0.765–0.916) achieved significantly higher AUCs and greater diagnostic accuracy than the conventional MRI model (AUC =0.771, accuracy =70.52%, sensitivity =62.50%, specificity =88.09%, 95% CI: 0.680–0.861). Survival analyses revealed that high T2-weighted imaging (T2WI) heterogeneity, a multilobulated or other tumor shape, pVol_Habitat 3 ≥26.25% and MK_Mean_Habitat 3 ≥0.913 were significantly associated with poor outcomes (all P<0.05).ConclusionsMulti-b-value DWI-based habitat imaging effectively characterizes ITH in STS and demonstrates significant clinical potential for the noninvasive preoperative prediction of histological grade and prognosis.Keywords: Soft tissue sarcoma (STS), intratumoral heterogeneity (ITH), habitat imaging, histological grade, prognosis Introduction Soft tissue sarcomas (STSs) are rare mesenchymal malignancies characterized by marked inter-subtype heterogeneity (1). According to the 2020 World Health Organization (WHO) classification, high-grade STSs, such as undifferentiated pleomorphic sarcoma (UPS) and rhabdomyosarcoma, commonly exhibit high intratumoral heterogeneity (ITH), while low-grade STSs, such as myxoid liposarcoma, show low ITH (2). ITH drives the formation of multiple distinct biological habitats in tumors, whose microscopic spatial distribution and composition influence tumor aggressiveness, therapeutic response, and prognosis (3). The Fédération Nationale des Centres de Lutte Contre le Cancer (FNCLCC) grading system plays a significant role in guiding clinical decision-making for STSs; however, its accuracy is often compromised by ITH, which introduces substantial uncertainty into preoperative tumor grading (4). Thus, the marked ITH of STSs represents a fundamental challenge in their clinical management. Core needle biopsy (CNB) remains the primary method for preoperative histologic diagnosis and grading of STSs; however, it is subject to inherent and difficult-to-overcome limitations (5). As an invasive procedure, CNB carries a theoretical risk of needle-tract seeding and is constrained by patient tolerance (6,7). More critically, due to its limited and random sampling, CNB frequently fails to capture the most aggressive and high-activity tumor regions, leading to systematic underestimation of tumor grade (8). This may lead to undertreatment, such as the failure to provide neoadjuvant therapy to high-risk patients, thereby compromising local control, increasing metastatic risk, and worsening prognosis (6). Thus, there is an urgent clinical need to develop a noninvasive approach capable of comprehensively mapping ITH for the accurate preoperative prediction of histological grade and prognosis, thereby optimizing the personalized management of STSs. Habitat imaging is an advanced imaging analysis technique based on unsupervised clustering that automatically identifies and segments functionally distinct tumor subregions according to voxel-wise feature similarity, thereby enabling intuitive spatial mapping of ITH (9,10). This approach has demonstrated value in characterizing the tumor microenvironment of various malignancies, including hepatocellular carcinoma (11), glioma (12), and breast cancer (13). Previous methods for assessing ITH, including histogram analysis and conventional radiomics, rely on global features that fail to capture subregional spatial information and often lack biological interpretability (14,15). In contrast, multi-b-value diffusion-weighted imaging (DWI)-based habitat imaging provides an interpretable, biologically principled framework for evaluating three dimensions of ITH: cellularity, perfusion, and microstructural complexity. The intravoxel incoherent motion (IVIM) model separately quantifies the true diffusion coefficient (D), which reflects the degree of free water diffusion, and the perfusion fraction (f), which represents microcapillary network perfusion (16). Mean kurtosis coefficient (MK), derived from diffusion kurtosis imaging (DKI), measures microstructural heterogeneity and complexity (17). Thus, we hypothesized that IVIM- and DKI-based habitat imaging could visually and quantitatively assess ITH in STSs. This study aimed to preoperatively and noninvasively predict the histological grade and prognosis of STS using multi-b-value DWI-based habitat imaging. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0694/rc). Methods Study population The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of The Second Hospital of Dalian Medical University (No. KY2025-069-01). Written informed consent was obtained from all patients. Between January 2018 and March 2025, 377 patients with clinically suspected soft tissue tumors were initially recruited from The Second Hospital of Dalian Medical University. The inclusion criteria were as follows: (I) potentially resectable soft tissue mass; (II) age ≥18 years; and (III) undergoing magnetic resonance imaging (MRI). The exclusion criteria were as follows: (I) pathology-confirmed non-STS; (II) a history of preoperative neoadjuvant therapy; (III) inadequate or poor IVIM/DKI image quality; (IV) unavailable histologic grading information; (V) refusal of further follow-up; and/or (VI) follow-up duration <6 months. Ultimately, 101 participants were included in the study (Figure 1). Flow diagram of the patient inclusion process. DKI, diffusion kurtosis imaging; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; IVIM, intravoxel incoherent motion; LI, labeling index; MRI, magnetic resonance imaging. MRI examination protocol All examinations were performed on a 3.0-Tesla whole-body scanner (Discovery MR750w, GE Healthcare, Waukesha, WI, USA) using appropriate transmit-receive coils. The protocol included fast spin-echo T1-weighted imaging (T1WI) [repetition time (TR)/echo time (TE), 478–836/14–18 ms; slice thickness, 3 mm] and fat-suppressed T2-weighted imaging (T2WI) (TR/TE, 3,042–5,521/70–82 ms; slice thickness, 3 mm). For IVIM imaging, the parameters were as follows: TR/TE, 3,000/70 ms; b-values, 0, 25, 50, 75, 100, 200, 500, 800, 1,000, 1,200, and 1,500 s/mm2; number of excitations (NEX), 2–4; matrix =128×128; acquisition time, approximately 8 min 20 s. For DKI, the parameters were as follows: TR/TE, 5,000/80 ms; b-values, 0, 1,000, and 2,000 s/mm2; 15 directions per b-value; NEX, 4; matrix, 128×128; acquisition time, 8 min 50 s. MRI image analysis Image processing and tumor segmentation Images were processed using the integrated Functool software on GE ADW 4.7 workstation. For IVIM analysis, the mean apparent diffusion coefficient (MADC) module applied a biexponential model to multi-b-value images to generate parametric maps of D and f. D reflects pure water molecular diffusion, while f represents microcirculatory perfusion. For DKI analysis, the dedicated DKI module employed a kurtosis model to high-b-value images to generate MK maps, characterizing tissue microstructural complexity. All parametric maps were then co-registered and resampled to axial T2WI space using the Advanced Normalization Tools (ANTs) toolkit implemented in Python for spatial normalization. Two musculoskeletal radiologists from a multidisciplinary sarcoma center (with 14 and 6 years of experience, respectively), who were blinded to the pathological results, independently reviewed the MRI studies. Compared with DWI, T2WI provides higher spatial resolution, clearer anatomical delineation, and better resistance to image deformation. Therefore, with reference to other MRI sequences, the entire STS lesion was manually delineated on each axial T2WI slice using ITK‑SNAP software to create a corresponding volume of interest (VOI). A final consensus VOI was established between the two readers and subsequently reviewed and confirmed by a third senior radiologist (with more than 20 years of musculoskeletal imaging experience). The consensus VOI was then automatically transferred to all co-registered parametric maps for subsequent analysis. Habitat imaging and feature extraction Habitat imaging was performed by applying K-means clustering analysis to the D, f, and MK parametric maps using Python software in the Visual Studio Code (VS Code) environment. All tumor voxels from the segmented STS regions across the entire patient cohort were first consolidated into an N × 3 matrix, where N represents the total number of voxels, and the three columns correspond to the DWI-derived parametric maps (D, f, and MK). In this unsupervised framework, K is a prespecified integer representing the number of habitat subregions to be generated. A centroid represents the representative point of a cluster and is defined as the vector of mean D, f, and MK values of all voxels assigned to that cluster. The initial centroids were randomly selected and then iteratively updated to the optimal cluster center. Inertia was used to quantify cluster compactness and was calculated as the sum of squared distances from all voxels to their respective centroids. After initializing K centroids, the algorithm iteratively assigned each voxel to the nearest centroid based on Euclidean distance and then recalculated the centroids according to the new mean values of the assigned voxels; this process was repeated until convergence to minimize inertia. To determine the optimal K value, the elbow method was used to plot inertia against an increasing range of K values and identify the inflection point at which the rate of inertia reduction sharply decreased, thereby indicating the optimal number of habitats that provided the best balance between within-cluster cohesion and between-cluster separation. The silhouette coefficient was additionally computed to assess the stability of the partition. Once the optimal K value was determined, quantitative features were extracted from each defined habitat, including volume, volume percentage (pVol), and the mean values of D (D_Mean), f (f_Mean), and MK (MK_Mean). Tumors can be conceptualized as complex ecosystems composed of distinct cellular subpopulations, with regional differences in cell density, microvascular perfusion, and microstructural complexity. Through unsupervised K-means clustering, habitat imaging partitions tumors into multiple voxel-based clusters with similar functional characteristics, enabling quantitative characterization of distinct intratumoral functional regions. The volume of each habitat subregion was defined as the number of voxels in the cluster multiplied by the physical volume of a single voxel. The pVol was calculated as the volume of a given habitat subregion divided by the total tumor volume, reflecting the relative proportion of a specific biological phenotype in the overall tumor. Both volume and pVol quantitatively represent the spatial extent occupied by different biological phenotypes in the tumor. Previous studies have demonstrated the value of habitat volume and pVol in characterizing tumor heterogeneity and predicting histological grade and prognosis (12,18). Consequently, the volume and pVol from habitat imaging may serve as quantitative indicators of tumor ecosystem architecture, reflecting the distributional proportion and distinctive traits of different habitats in the tumor. Pathological analysis Two pathologists (with 18 and 4 years of experience in musculoskeletal pathology, respectively), who were blinded to clinical and radiological information, independently evaluated the surgical specimens. Histopathological diagnoses were established according to the 2020 WHO classification of STSs. Tumor grading was performed using the FNCLCC system based on the assessment of tumor differentiation, necrosis, and mitotic count, with tumors classified as Grade I, II, or III (19). Immunohistochemical analysis of Ki-67 expression was performed by calculating the percentage of positive nuclei in hotspot regions at ×400 magnification. Cases with Ki-67 expression<30% were categorized as the low-expression group, while those with Ki-67 ≥30% were categorized as the high-expression group (20). For final risk stratification, the modified grading method was applied: the low-grade group comprised tumors with FNCLCC Grade I or Grade II and Ki-67 low-expression, while the high-grade group comprised tumors with FNCLCC Grade III or Grade II and Ki-67 high-expression (21). Follow-up Overall survival (OS) and metastasis-free survival Patient follow-up included clinical physical examinations and assessments using computed tomography or MRI. Follow-up was conducted every 3 months for the first 2 years, and every 6 months thereafter. OS was defined as the interval from the date of initial pathological diagnosis to death from any cause. Metastasis-free survival (MFS) was defined as the interval from treatment initiation to the occurrence of distant metastasis or death from any cause, whichever occurred first, with the date of primary surgery serving as the time origin in this study. Statistical analysis The statistical analysis was performed using SPSS and GraphPad Prism software. Continuous variables were presented as mean ± standard deviation (SD), while categorical variables were presented as frequencies and percentages. Inter‑observer agreement for the assessment of morphological features was evaluated using Cohen’s kappa coefficient. Differences between the low‑ and high‑grade groups were compared using the Chi‑squared test, independent‑samples t‑test, or Mann-Whitney U test, as appropriate. Univariable and multivariable logistic regression analyses were performed to identify the factors associated with histological grade. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, and model calibration was assessed using the Hosmer-Lemeshow test. Kaplan-Meier analysis was used to identify factors associated with OS and MFS. Cox proportional hazards regression analysis was then performed to calculate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). Results Participants characteristics This prospective study enrolled 101 patients with STS. Preoperative clinical features, pathological features derived from surgical specimens, stratified according to the FNCLCC grade incorporating Ki-67 expression, are summarized in Table 1. The patients had a mean age of 61.22±16.41 years and most were male (53.47%). The most common tumor location was the thigh (45.54%). The Ki-67 labeling index (LI) ranged from 5% to 90%, with a mean ± SD of 35.64%±22.52%. The most common histological subtypes were UPS (17.82%), myxoid liposarcoma (14.85%), and myxofibrosarcoma (14.85%). According to the FNCLCC grading system, the 101 tumors were classified as Grade I (n=21), II (n=32), and III (n=48). Among the Grade II tumors, 21 showed low Ki-67 expression, while 11 showed high expression. Using the modified grading method, 42 lesions were ultimately categorized into the low-grade group and 59 into the high-grade group. Table 1Clinical and pathological features of STS patients stratified by FNCLCC grade incorporating Ki-67 expressionCharacteristicPatients (n=101)Grade I (n=21)Grade II (n=32)Grade III (n=48)Ki-67 LI <30% (n=21)Ki-67 LI ≥30% (n=11)Age (years)61.22±16.4158.67±14.3459.62±18.6364.45±12.6062.29±17.21Sex Male54 (53.47)6 (28.57)13 (61.90)6 (54.55)29 (60.42) Female47 (46.53)15 (71.43)8 (38.10)5 (45.45)19 (39.58)Location Upper arm3 (2.97)1 (4.76)0 (0.00)0 (0.00)2 (4.17) Forearm11 (10.89)1 (4.76)2 (9.52)3 (27.27)5 (10.42) Thigh46 (45.54)13 (61.90)7 (33.33)3 (27.27)23 (47.92) Lower leg9 (8.91)2 (9.52)3 (14.29)2 (18.18)2 (4.17) Trunk32 (31.68)4 (19.05)9 (42.86)3 (27.27)16 (33.33)Depth Superficial23 (22.77)11 (52.38)8 (38.10)2 (18.18)2 (4.17) Deep66 (65.35)9 (42.86)11 (52.38)8 (72.73)38 (79.17) Superficial + deep12 (11.88)1 (4.76)2 (9.52)1 (9.10)8 (16.66)Ki-67 LI (%)35.64±22.5210.95±7.6818.81±4.7236.36±11.4253.64±16.94Histologic type Dermatofibrosarcoma protuberans5 (4.95)2 (9.52)3 (14.29)0 (0.00)0 (0.00) Fibrosarcoma11 (10.89)2 (9.52)2 (9.52)2 (18.18)5 (10.42) Myxofibrosarcoma15 (14.85)6 (28.57)6 (28.57)2 (18.18)1 (2.09) Dedifferentiated liposarcoma3 (2.97)0 (0.00)0 (0.00)0 (0.00)3 (6.25) Pleomorphic liposarcoma3 (2.97)0 (0.00)0 (0.00)0 (0.00)3 (6.26) Myxoid liposarcoma15 (14.85)9 (42.86)4 (19.05)2 (18.18)0 (0.00) Leiomyosarcoma12 (11.88)0 (0.00)0 (0.00)1 (5.56)11 (22.92) Synovial sarcoma6 (5.94)0 (0.00)4 (19.05)1 (5.56)1 (2.09) Rhabdomyosarcoma3 (2.97)0 (0.00)0 (0.00)0 (0.00)3 (6.25) MGCTS1 (0.99)0 (0.00)1 (4.76)0 (0.00)0 (0.00) MPNST5 (4.95)0 (0.00)1 (4.76)0 (0.00)4 (8.33) UPS18 (17.82)0 (0.00)0 (0.00)3 (27.27)15 (31.25) Alveolar soft part sarcoma1 (0.99)0 (0.00)0 (0.00)0 (0.00)1 (2.09) Clear cell sarcoma1 (0.99)0 (0.00)0 (0.00)0 (0.00)1 (2.09) Low-grade malignant myofibroblastic sarcoma1 (0.99)1 (4.76)0 (0.00)0 (0.00)0 (0.00) Extraskeletal chondrosarcoma1 (0.99)1 (4.76)0 (0.00)0 (0.00)0 (0.00)Data are presented as mean ± standard deviation or n (%). FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; LI, labeling index; MGCTS, malignant giant cell tumor of tendon sheath; MPNST, malignant peripheral nerve sheath tumor; STS, soft tissue sarcoma; UPS, undifferentiated pleomorphic sarcoma. Clinical and pathological features of STS patients stratified by FNCLCC grade incorporating Ki-67 expression Data are presented as mean ± standard deviation or n (%). FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; LI, labeling index; MGCTS, malignant giant cell tumor of tendon sheath; MPNST, malignant peripheral nerve sheath tumor; STS, soft tissue sarcoma; UPS, undifferentiated pleomorphic sarcoma. Conventional MRI features Interobserver agreement for all assessed MRI features ranged from moderate to excellent (kappa =0.59–0.83). Four conventional MRI features showed significant differences between the low- and high-grade groups: tumor size, tumor shape, T2WI heterogeneity, and peritumoral enhancement (P<0.05). No significant differences were observed between groups in terms of tumor margin, necrosis, T1WI heterogeneity, heterogeneity on post-gadolinium T1WI, peritumoral edema, tail sign, or bone, nerve and vessel involvement (Table 2). Table 2Conventional MRI features of low- and high-grade STSs according to the modified grading methodFeaturesLow-grade (n=42)High-grade (n=59)P valueTumor size (cm)6.41±3.758.99±5.340.008*Tumor shape0.003* Ovoid33 (78.57)29 (49.15) Multilobulated or other9 (21.43)30 (50.85)Tumor margin0.428 Well-defined24 (57.14)29 (49.15) Poorly defined18 (42.85)30 (50.85)Tumor necrosis0.317 <50%26 (61.90)35 (59.32) ≥50%16 (38.10)24 (40.68)T1WI heterogeneity0.403 <50%33 (78.57)42 (71.19) ≥50%9 (21.43)17 (28.81)T2WI heterogeneity0.002* <50%26 (61.90)18 (30.51) ≥50%16 (38.10)41 (69.50)T1WI after gadolinium injection heterogeneity0.225 <50%20 (47.62)21 (35.59) ≥50%22 (52.38)38 (64.41)Peritumoral edema0.460 No18 (42.86)21 (35.59) Yes24 (57.14)38 (64.41)Peritumoral enhancement0.045* No33 (78.57)26 (47.27) Yes9 (21.43)33 (52.73)Tail sign0.358 No32 (76.19)40 (67.80) Yes10 (23.81)19 (32.20)Bone, nerve, vessel involvement0.350 No39 (92.86)50 (84.75) Yes3 (7.14)9 (15.25)Data are presented as mean ± standard deviation or n (%). *, P<0.05. MRI, magnetic resonance imaging; STS, soft tissue sarcoma; T1WI, T1-weighted imaging; T2WI, T2-weighted imaging. Conventional MRI features of low- and high-grade STSs according to the modified grading method Data are presented as mean ± standard deviation or n (%). *, P<0.05. MRI, magnetic resonance imaging; STS, soft tissue sarcoma; T1WI, T1-weighted imaging; T2WI, T2-weighted imaging. Habitat features In this study, the optimal number of habitats was determined to be 4 (K=4) based on the combined results of the elbow method and silhouette coefficient analysis. The elbow point, defined as the inflection point where the rate of decrease in the within-cluster sum of squared errors changed from steep to gradual, corresponded to K=4 (Figure 2A). Concurrently, the highest silhouette coefficient (0.581) was also observed at K=4 (Figure 2B). The quantitative parameter distributions of the four identified habitats are detailed in Figure 2C. Habitat 1 exhibited a D value of (1.23±0.29) ×10⁻3 mm2/s, an f value of 0.27±0.05, and an MK value of 0.74±0.27. Habitat 2 exhibited a D value of (1.11±0.41) ×10⁻3 mm2/s, an f value of 0.25±0.09, and an MK value of 0.81±0.33. Habitat 3 exhibited a D value of (0.96±0.17) ×10⁻3 mm2/s, an f value of 0.22±0.08, and an MK value of 0.91±0.12. Habitat 4 exhibited a D value of (1.71±0.61) ×10⁻3 mm2/s, an f value of 0.19±0.08, and an MK value of 0.46±0.23. Representative habitat images of STSs are presented in Figure 3. Habitat encoding and characteristics. The elbow method (A) and silhouette score (B) both determined the optimal number of habitats, which was four. (C) Quantitative characteristics of the four habitats. D, true diffusion coefficient (×10–3 mm2/s); f, perfusion fraction; MK, mean kurtosis coefficient; WCSS, within-cluster sum of squares. Representative habitat images of different histological grades. Immunohistochemical analysis of Ki-67 expression was performed by calculating the percentage of positive nuclei in hotspot regions at ×400 magnification. Grade I tumor from the low-grade group (A), myxoid liposarcoma, with a Ki-67 LI of 5%. Grade II tumor from the low-grade group (B), myxofibrosarcoma, with a Ki-67 LI of 25%. Grade II tumor from the high-grade group (C), undifferentiated pleomorphic sarcoma, with a Ki-67 LI of 70%. Grade III tumor from the high-grade group (D), leiomyosarcoma, with a Ki-67 LI of 80%. LI, labeling index; T1WI+C, T1-weighted imaging with contrast enhancement; T2WI, T2-weighted imaging. Significant intergroup differences were observed in specific habitat features (Table 3). Compared with the low-grade group, the high-grade group exhibited lower pVol (33.60%±7.23% vs. 21.49%±10.59%, P<0.001) and lower D_Mean [1.29±0.24 vs. 1.17±0.24 (×10⁻3 mm2/s), P=0.044], but higher f_Mean (0.25±0.04 vs. 0.28±0.05, P=0.002) in Habitat 1. Conversely, in Habitat 3, the high-grade group demonstrated higher pVol (29.64%±6.89% vs. 21.48%±7.03%, P<0.001) and MK_Mean (0.98±0.11 vs. 0.81±0.07, P<0.001), but lower D_Mean [0.89±0.18 vs. 1.06±0.08 (×10⁻3 mm2/s), P<0.001]. No other habitat features differed significantly between the two groups. Table 3Habitat features of low- and high-grade STSsFeaturesLow-grade (n=42)High-grade (n=59)P valueHabitat 1 pVol (%)33.60±7.2321.49±10.59<0.001* Volume, mm332,227±41,66619,043±26,3380.054 D_Mean, ×10−3 mm2/s1.29±0.241.17±0.240.044* f_Mean0.25±0.040.28±0.050.002* MK_Mean0.68±0.270.78±0.280.090Habitat 2 pVol (%)22.81±12.6123.25±12.590.863 Volume, mm320,026±43,32022,691±41,1100.754 D_Mean, ×10−3 mm2/s1.16±0.371.08±0.440.336 f_Mean0.24±0.100.26±0.090.344 MK_Mean0.74±0.190.86±0.400.069Habitat 3 pVol (%)21.48±7.0329.64±6.89<0.001* Volume, mm317,241±23,49128,957±45,3810.129 D_Mean, ×10−3 mm2/s1.06±0.080.89±0.18<0.001* f_Mean0.23±0.100.22±0.070.405 MK_Mean0.81±0.070.98±0.11<0.001*Habitat 4 pVol (%)22.34±12.1425.36±12.320.224 Volume, mm320,229±24,61723,760±40,3860.616 D_Mean, ×10−3 mm2/s1.68±0.581.73±0.640.675 f_Mean0.19±0.090.19±0.080.926 MK_Mean0.44±0.190.48±0.260.432Data are presented as mean ± standard deviation. *, a statistical difference (P<0.05). D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; MK, mean kurtosis coefficient; MK_Mean, mean MK value; pVol, volume percentage; STS, soft tissue sarcoma. Habitat features of low- and high-grade STSs Data are presented as mean ± standard deviation. *, a statistical difference (P<0.05). D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; MK, mean kurtosis coefficient; MK_Mean, mean MK value; pVol, volume percentage; STS, soft tissue sarcoma. Model construction and performance evaluation for grade prediction models Univariate logistic regression analyses were initially performed on clinical features, conventional MRI features, and habitat features separately to identify predictors of histological grade. Significant predictors from each set were subsequently entered in multivariable analyses, leading to the construction of three distinct models: the conventional MRI model, the habitat model, and the combined model (Table 4). The goodness-of-fit of all models was confirmed using the Hosmer-Lemeshow test (all P>0.05). The ROC curves for the three models in predicting histological grade are presented in Figure 4. The combined model demonstrated the best diagnostic performance and accuracy, with an area under the curve (AUC) of 0.896 (95% CI: 0.837–0.916; sensitivity =79.70%; specificity =80.95%; accuracy =87.28%). The habitat model achieved an AUC of 0.841 (95% CI: 0.765–0.916; sensitivity =69.50%; specificity =85.71%; accuracy =76.75%). The conventional MRI model had an AUC of 0.771 (95% CI: 0.680–0.861; sensitivity =62.50%; specificity =88.09%; accuracy =70.52%). Table 4Univariable and multivariable logistic regression analyses for histological grade prediction modelsModelUnivariableMultivariableModel fitOR (95% CI)POR (95% CI)PClinical model Age1.012 (0.989–1.038)0.285 Sex0.404 (0.180–0.910)0.290 Tumor location1.222 (1.163–1.316)0.555 Tumor depth1.333 (1.191–1.931)0.564Conventional MRI model0.940 Tumor size1.136 (1.028–1.256)0.012*1.164 (1.039–1.304)0.009* Tumor shape0.302 (0.126–0.724)0.007*0.308 (0.112–0.845)0.002* T2WI heterogeneity0.299 (0.130–0.684)0.004*0.223 (0.056–0.576)0.002* Peritumoral enhancement0.408 (0.181–0.923)0.031*0.676 (0.258–1.770)0.425Habitat model0.552 pVol_Habitat 10.871 (0.823–0.923)<0.001*0.676 (0.548–0.852)0.404 D_Mean_Habitat 10.999 (0.997–1.000)0.051 f_Mean_Habitat 11.001 (1.000–1.002)0.005*1.001 (0.999–1.003)0.186 pVol_Habitat 31.185 (1.099–1.277)<0.001*1.057 (0.928–1.203)<0.001* D_Mean_Habitat 30.989 (0.984–0.994)<0.001*0.993 (0.986–1.001)0.083 MK_Mean_Habitat 31.021 (1.012–1.030)<0.001*1.027 (1.010–1.045)0.002*Combined model0.646 Tumor size1.136 (1.028–1.256)0.012*1.004 (0.805–1.252)0.975 Tumor shape0.302 (0.126–0.724)0.007*0.368 (0.116–0.923)0.026* T2WI heterogeneity0.302 (0.126–0.724)0.004*0.153 (0.011–2.163)0.165 pVol_Habitat 31.185 (1.099–1.277)<0.001*1.207 (1.091–1.277)<0.001* MK_Mean_Habitat 31.021 (1.012–1.030)<0.001*1.033 (1.015–1.051)<0.001*The calibration of the logistic regression model was assessed using the Hosmer-Lemeshow goodness-of-fit test, and a model with P>0.05 was considered to have good fit. *, P<0.05. CI, confidence interval; D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; MK, mean kurtosis coefficient; MK_Mean, mean MK value; MRI, magnetic resonance imaging; OR, odds ratio; pVol, volume percentage; T2WI, T2-weighted imaging. Univariable and multivariable logistic regression analyses for histological grade prediction models The calibration of the logistic regression model was assessed using the Hosmer-Lemeshow goodness-of-fit test, and a model with P>0.05 was considered to have good fit. *, P<0.05. CI, confidence interval; D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; MK, mean kurtosis coefficient; MK_Mean, mean MK value; MRI, magnetic resonance imaging; OR, odds ratio; pVol, volume percentage; T2WI, T2-weighted imaging. Model performance for distinguishing between low- and high-grade STS. AUC, area under the curve; CI, confidence interval; MRI, magnetic resonance imaging; STS, soft tissue sarcoma. Analysis of habitat features and Ki-67 expression between myxoid STS and UPS An analysis was conducted in patients with myxoid STS (including myxofibrosarcoma and myxoid liposarcoma; n=30) and UPS (n=18) within the study cohort. Significant differences in habitat features and Ki-67 expression were observed between myxoid STS and UPS. Compared with myxoid STS, UPS exhibited higher pVol (29.12%±5.70% vs. 25.34%±6.13%, P=0.039) and f (0.29±0.06 vs. 0.26±0.02, P=0.034) in Habitat 1, as well as lower pVol (20.43%±9.50% vs. 28.30%±12.91%, P=0.040) and D [1.58±0.43 vs. 1.83±0.31 (×10⁻3 mm2/s), P=0.033] in Habitat 4. Further, Ki-67 expression differed significantly between UPS and myxoid STS, with UPS demonstrating markedly higher Ki-67 expression (51.67%±15.43% vs. 27.17%±20.33%, P<0.001) (Table 5). Table 5Analysis of habitat features and Ki-67 expression between myxoid STS and UPSFeaturesMyxoid STS (n = 30)UPS (n =18)P valueHabitat 1 pVol (%)25.34±6.1329.12±5.700.039* Volume, mm32,4731±58562,8421±58560.298 D_Mean, ×10−3 mm2/s1.19±0.491.06±0.450.380 f_Mean0.26±0.020.29±0.060.034* MK_Mean0.76±0.150.80±0.220.473Habitat 2 pVol (%)23.87±13.4722.03±11.440.374 Volume, mm323,297±13,66421,501±11,1650.630 D_Mean, ×10−3 mm2/s1.12±0.050.99±0.290.321 f_Mean0.24±0.100.26±0.090.598 MK_Mean0.74±0.190.86±0.400.620Habitat 3 pVol (%)21.97±11.4228.40±14.120.091 Volume, mm321,442±11,14527,718±13,6640.192 D_Mean, ×10−3 mm2/s1.04±0.560.81±0.0.280.115 f_Mean0.22±0.070.22±0.100.905 MK_Mean0.97±0.260.99±0.190.810Habitat 4 pVol (%)28.30±12.9120.43±9.500.040* Volume, mm327,620±12,60019,939±9,2720.567 D_Mean, ×10−3 mm2/s1.83±0.311.58±0.430.033* f_Mean0.18±0.060.19±0.040.541 MK_Mean0.49±0.190.46±0.150.571Ki-67 LI (%)27.17±20.3351.67±15.43<0.001*Data are presented as mean ± standard deviation. Myxoid STS: myxofibrosarcoma, myxoid liposarcoma. *, a statistically significant difference (P<0.05). D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; LI, labeling index; MK, mean kurtosis coefficient; MK_Mean, mean MK value; pVol, volume percentage; STS, soft tissue sarcoma; UPS, undifferentiated pleomorphic sarcoma. Analysis of habitat features and Ki-67 expression between myxoid STS and UPS Data are presented as mean ± standard deviation. Myxoid STS: myxofibrosarcoma, myxoid liposarcoma. *, a statistically significant difference (P<0.05). D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; LI, labeling index; MK, mean kurtosis coefficient; MK_Mean, mean MK value; pVol, volume percentage; STS, soft tissue sarcoma; UPS, undifferentiated pleomorphic sarcoma. Survival analysis The follow-up duration for the cohort ranged from 6 to 87 months. The OS and MFS rates were 60.39% and 54.45%, respectively. The survival analysis results are summarized in Tables 6,7 and Figure 5. Multivariable Cox proportional hazards regression analysis identified five independent predictors associated with OS: multilobulated or other tumor shape (HR =1.328, 95% CI: 1.144–1.751, P=0.018), T2WI heterogeneity ≥50% (HR =1.270, 95% CI: 1.083–1.883, P=0.030), pVol_Habitat 3 ≥26.25% (HR =1.295, 95% CI: 1.102–1.521, P=0.002), MK_Mean_Habitat 3 ≥0.913 (HR =1.005, 95% CI: 0.995–1.009, P=0.044), and FNCLCC grade (HR =1.325, 95% CI: 1.298–2.284, P=0.007). Multivariable Cox proportional hazards regression analysis identified three independent predictors associated with MFS: T2WI heterogeneity ≥50% (HR =1.216, 95% CI: 1.059–1.781, P=0.019), pVol_Habitat 3 ≥26.25% (HR =1.178, 95% CI: 1.046–1.328, P=0.007), and FNCLCC grade (HR =1.360, 95% CI: 1.068–2.170, P=0.004). Table 6Multivariable Cox regression analysis of overall survivalFeaturesHR (95% CI)P valueTumor size (cm) <7.9231.00 (reference) ≥7.9231.044 (0.950–1.148)0.369Tumor shape Ovoid1.00 (reference) Multilobulated or other1.328 (1.144–1.751)0.018*T2WI heterogeneity <50%1.00 (reference) ≥50%1.270 (1.083–1.883)0.030*Peritumoral enhancement Absent1.00 (reference) Present1.200 (1.051–1.784)0.210pVol_ Habitat 1 (%) <26.5271.00 (reference) ≥26.5271.011 (0.963–1.061)0.658D_Mean_Habitat 1 (×10−3 mm2/s) <1.2251.00 (reference) ≥1.2250.999 (0.996–1.000)0.144f_Mean_Habitat 1 <0.2701.00 (reference) ≥0.2701.001 (0.999–1.003)0.251pVol_Habitat 3 (%) <26.2551.00 (reference) ≥26.2551.295 (1.102–1.521)0.002*D_Mean_Habitat 3 (×10−3 mm2/s) <0.9611.00 (reference) ≥0.9610.998 (0.997–1.002)0.858MK_Mean_Habitat 3 <0.9131.00 (reference) ≥0.9131.005 (0.995–1.009)0.044*FNCLCC grade I1.00 (reference) II1.104 (1.003–1.724)0.063 III1.325 (1.298–2.284)0.007*Histologic subtype Non‑myxoid STS1.00 (reference) Myxoid STS0.738 (0.227–0.973)0.613*, a statistically significant difference (P<0.05). CI, confidence interval; D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; HR, hazard ratio; MK, mean kurtosis coefficient; MK_Mean, mean MK value; pVol, volume percentage; STS, soft tissue sarcoma. Multivariable Cox regression analysis of overall survival *, a statistically significant difference (P<0.05). CI, confidence interval; D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; HR, hazard ratio; MK, mean kurtosis coefficient; MK_Mean, mean MK value; pVol, volume percentage; STS, soft tissue sarcoma. Table 7Multivariable Cox regression analysis of metastasis-free survivalFeaturesHR (95% CI)P valueTumor size (cm) <7.9231.00 (reference) ≥7.9231.060 (0.969–1.159)0.208Tumor shape Ovoid1.00 (reference) Multilobulated or other1.053 (1.023–1.156)0.111T2WI heterogeneity <50%1.00 (reference) ≥50%1.216 (1.059–1.781)0.019*Peritumoral enhancement Absent1.00 (reference) Present1.234 (1.066–1.836)0.316pVol_ Habitat 1 (%) <26.5271.00 (reference) ≥26.5271.023 (0.979–1.069)0.125D_Mean_Habitat 1 (×10−3 mm2/s) <1.2251.00 (reference) ≥1.2250.999 (0.998–1.000)0.063f_Mean_Habitat 1 <0.2701.00 (reference) ≥0.2701.001 (1.000–1.003)0.123pVol_Habitat 3 (%) <26.2551.00 (reference) ≥26.2551.178 (1.046–1.328)0.007*D_Mean_Habitat 3 (×10−3 mm2/s) <0.9611.00 (reference) ≥0.9610.999 (0.997–1.002)0.492MK_Mean_Habitat 3 <0.9131.00 (reference) ≥0.9130.999 (0.994–1.005)0.138FNCLCC grade I1.00 (reference) II1.007 (1.001–1.289)0.071 III1.360 (1.068–2.170)0.004*Histologic subtype Non‑myxoid STS1.00 (reference) Myxoid STS0.535 (0.167–1.711)0.291*, a statistically significant difference (P<0.05). CI, confidence interval; D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; HR, hazard ratio; MK, mean kurtosis coefficient; pVol, volume percentage; MK_Mean, mean MK value; STS, Soft tissue sarcoma; T2WI, T2-weighted imaging. Multivariable Cox regression analysis of metastasis-free survival *, a statistically significant difference (P<0.05). CI, confidence interval; D, true diffusion coefficient; D_Mean, mean D value; f, perfusion fraction; f_Mean, mean f value; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; HR, hazard ratio; MK, mean kurtosis coefficient; pVol, volume percentage; MK_Mean, mean MK value; STS, Soft tissue sarcoma; T2WI, T2-weighted imaging. Survival analysis of OS and MFS in patients with STS. Graphs showing Kaplan-Meier survival curves for OS according to the features of STS: tumor shape (A), T2WI heterogeneity (B), pVol_Habitat 3 (C), MK_Mean_Habitat 3 (D), and FNCLCC grade (E). Graphs showing Kaplan-Meier survival curves for MFS according to the features of STS: T2WI heterogeneity (F), pVol_Habitat 3 (G), and FNCLCC grade (H). FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; MFS, metastasis-free survival; MK, mean kurtosis coefficient; MK_Mean, mean MK value; OS, overall survival; pVol, volume percentage; STS, soft tissue sarcoma; T2WI, T2-weighted imaging. Discussion This study presents a novel approach using multi-b-value DWI-based habitat imaging for the preoperative non‑invasive prediction of histological grade and prognosis in STSs. By integrating parametric maps (D, f, and MK), this approach enables the quantification and visualization of ITH from three complementary dimensions: water molecule diffusion, microvascular perfusion, and microstructural complexity. We adopted a modified histological grading method incorporating Ki‑67 expression and constructed three predictive models using logistic regression based on conventional MRI and habitat features. The results demonstrated that both the combined model and habitat model achieved significantly higher AUCs and greater diagnostic accuracy than the conventional MRI model. Moreover, habitat features showed significant prognostic value. Therefore, multi-b-value DWI-based habitat imaging may provide substantial clinical benefits by enabling more accurate preoperative grading and prognostic assessment to guide decision-making for patients with STSs. ITH is a fundamental biological feature of STSs that contributes to sampling errors and histological grade underestimation during preoperative CNB and is closely associated with poor prognosis (22). The key contribution of the habitat imaging approach employed in this study lies in its ability to transform the abstract concept of ITH into quantifiable, spatially localized imaging biomarkers with established pathophysiological correlates (23). Unsupervised clustering analysis delineated four distinct habitats, among which features derived from Habitats 1 and 3 exhibited significant differences between low- and high-grade STSs. Habitat 1, characterized by high perfusion, high cellularity, and moderate microstructural complexity, may represent a highly active angiogenic region. Habitat 3, characterized by very high cellularity, relatively low perfusion, and increased microstructural complexity, may represent a hypoxic region. This hypoxic microenvironment may induce epithelial-mesenchymal transition and upregulate metastasis-associated genes, thereby enhancing invasive and metastatic potential, which is consistent with previous evidence linking hypoxia to the prognosis of STSs (24,25). In the present study, high‑grade STSs showed a greater proportion of Habitat 3 than low‑grade STSs, and this feature was associated with OS and MFS, indicating that more aggressive tumors contain a greater spatial proportion of hypoxic regions, which may drive malignant progression and metastasis and is associated with poor prognosis. Therefore, Habitat 1 may provide nutrients that fuel tumor proliferation, while Habitat 3 may promote malignant evolution. Together, these habitats may form a functionally synergistic adverse cycle, driving STS progression and poor outcomes. In the FNCLCC system, Grade I and Grade III sarcomas are relatively well defined, while Grade II tumors often remain difficult to characterize (26). Previous studies have typically categorized Grade I as low-grade and Grade II–III as high-grade (27-29). However, this broad classification may obscure potential differences in biological behavior and clinical outcomes among histological subtypes in Grade II. Ki-67, a reliable marker of cellular proliferation, has been established as an independent predictor of histologic grade and prognosis in STSs. For more precise risk stratification, we adopted the modified grading system proposed by Li et al., which integrates Ki-67 expression with FNCLCC grade to establish a combined morphological-proliferative grading method (21). Under this refined criterion, logistic regression analysis showed that quantitative habitat imaging features were robust predictors of histological grade in STSs. The combined model demonstrated the highest diagnostic performance (AUC =0.896), while the habitat model (AUC =0.841) significantly outperformed the conventional MRI model (AUC =0.771). These findings indicate that habitat imaging provides independent and superior incremental diagnostic value over conventional morphological assessment. In this study, conventional MRI macroscopic markers, such as lobulated morphology and high T2WI heterogeneity, were reaffirmed as independent risk factors for poor outcomes, consistent with previous reports (21,30,31). Notably, pVol_Habitat 3 ≥26.255% and MK_Mean_Habitat 3 ≥0.913 emerged as significant microscopic predictors of poor outcomes. Additionally, higher pVol_Habitat 3 and greater microstructural complexity were directly associated with high-grade STSs. These findings establish an integrative link among ITH, histological grade, and prognosis via habitat imaging, demonstrating that this approach provides additional information reflecting the tumor microenvironment and underlying pathobiological processes beyond conventional morphological assessment, thereby enabling more precise risk stratification in STSs. Notably, myxoid STSs, including myxoid liposarcoma and myxofibrosarcoma, and UPS differ significantly in terms of perfusion and diffusion (32). Specifically, myxoid STSs are characterized by less restricted diffusion and lower perfusion, which may reflect their abundant myxoid stroma and relatively low cellularity, whereas UPS exhibits more restricted diffusion and higher perfusion. Given this considerable heterogeneity, we conducted separate subgroup analyses for these two groups. The results revealed significant differences between the two groups in terms of pVol and f in Habitat 1, pVol and D in Habitat 4, and Ki-67 expression. Compared with myxoid STSs, UPS exhibited a greater proportion of Habitat 1, with significantly higher perfusion in this region. In contrast, myxoid STSs showed a greater proportion of Habitat 4, with less restricted diffusion, which may be associated with the myxoid stromal component of these tumors. Further, Ki-67 expression differed significantly between UPS and myxoid STSs, with UPS demonstrating markedly higher Ki-67 expression, consistent with its more aggressive biological behavior characterized by enhanced tumor cell proliferation and higher malignant potential. Several limitations of this study should be noted. First, despite prospective enrollment over seven years, the cohort size remained relatively limited due to the rarity of STSs. Second, this was a single‑center study without external validation, which may limit the generalizability of the study findings. Third, the single‑TE IVIM model used in this study has an inherent limitation. This model assumes equal T2 relaxation times between tissue and vascular compartments and thus fails to account for their differential T2 (33,34). Because blood T2 is substantially longer than that of solid tissues, prolonged TE increases the signal weighting of the vascular compartment due to slower T2 decay, leading to overestimation of the f value. However, this represents a limitation of the model assumptions rather than an error in the equation itself. This limitation has been shown to be correctable. Simchick et al. applied post‑hoc R2 correction to single‑TE data and reported f estimates that differed from multi‑TE IVIM results by a mean absolute difference of ≤1.98%, with excellent repeatability (35). The multi‑TE extended T2‑IVIM model proposed by Jerome et al., which incorporates compartment‑specific T2 parameters, successfully separated the confounding T2 effects from the perfusion signal, effectively eliminated the TE dependence of f, and yielded a more accurate and reliable f value (36). Moreover, the coefficient of variation for single‑TE IVIM in routine practice is 4.9–7.0%, whereas that for multi‑TE IVIM can reach 23.4–35.0%, indicating that the multi‑TE model may sacrifice repeatability for theoretical precision (36). In contrast, the single‑TE model, with fewer parameters, offers practical advantages in repeatability. Therefore, our findings should be interpreted within the inherent assumptions of the single‑TE model. Future studies may consider multi‑TE acquisition or post‑hoc R2 correction for sensitivity analyses. Fourth, given the f values of 25–28% observed in this study, clarification is necessary to prevent conceptual misinterpretation arising from the numerical magnitude. The f value represents the relative signal contribution of microcirculatory blood flow to the total MRI signal in a voxel and is essentially a signal‑weighted fraction rather than an anatomical volume fraction. The f value is modulated by multiple factors, including T2 relaxation differences, TE, capillary geometry, blood flow velocity, and transcapillary water exchange rates. Jerome et al. reported regional f values of 22.4%±7% at TE =62 ms and 30.7%±11% at TE =102 ms (36) using a conventional single-TE IVIM model. In the present study, data acquired at TE =70 ms yielded f values of 25–28%, which fall within the reasonable range of the previously reported TE‑dependent values. Further, Stabinska et al. recently demonstrated that while f increased significantly with TE in the liver, no such TE dependence was observed in the renal cortex or medulla (37). Therefore, interpretation of f values should consider the specific physiological characteristics and T2 properties of STSs. Fifth, the clinical translation of this habitat imaging approach faces practical challenges: multi‑b‑value DWI acquisition prolongs scan time and subsequent post‑processing, including imaging registration and K‑means clustering, requires specialized software and computational resources, potentially restricting its use to academic or well‑equipped centers. Sixth, given the wide variety of histological STS subtypes, caution is warranted when interpreting findings from pooled analyses across subtypes with distinct biological behaviors. Conclusions This study demonstrates that multi-b-value DWI-based habitat imaging, by quantifying and visualizing ITH across multiple dimensions, provides an effective, noninvasive approach for the preoperative prediction of histological grade and prognosis in STSs. Using a modified grading method incorporating Ki-67 expression, we constructed prediction models for histological grade. Both the combined model and habitat model demonstrated superior diagnostic performance compared with the conventional MRI model, while the habitat features also provided significant prognostic information. This approach may address the limitations of CNB by supporting more precise preoperative risk stratification and individualized management of STSs. Supplementary The article’s supplementary files as Acknowledgments We would like to thank the Sarcoma Center of The Second Hospital of Dalian Medical University for establishing the research platform. 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. This study was approved by Institutional Ethics Committee of The Second Hospital of Dalian Medical University (No. KY2025-069-01). Written informed consent was obtained from all patients. Dalian Municipal Central Hospital was also informed of and agreed to the study. Footnotes Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0694/rc Funding: This study was supported by the National Natural Science Foundation of China (No. 82271975) and the Basic Research Project of the Education Department of Liaoning Province (No. LJ212510161041). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0694/coif). The authors have no conflicts of interest to declare. Data Sharing Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0694/dss References Associated Data Supplementary Materials The article’s supplementary files as Data Availability Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0694/dss