Added prognostic value of histogram features from preoperative multi-modal diffusion MRI in predicting Ki-67 proliferation for adult-type diffuse gliomas

Introduction

Despite great advances in the diagnosis and treatment of adult-type diffuse gliomas for decades, the overall clinical outcome varies in different patients (1). One of the major barriers in glioma treatment is the lack of therapeutic approaches based on histological or molecular classification, while gliomas are highly heterogeneous at the levels of genotype, molecule, and histology (2-4).

Ki-67, a marker of cellular proliferation, plays a crucial role in indicating the malignancy and prognosis of gliomas (5,6). High expression levels of Ki-67 labelling index (LI) indicate a higher proportion of cells in the division stage, leading to faster tumor growth, increased mutation rates, and a higher probability of developing drug resistance (6). This suggests that tumors with high Ki-67 LI may be more aggressive, thus resulting in poorer prognosis (7,8). Consequently, physicians can provide patients with more accurate prognosis information based on Ki-67 expression levels, enabling patients and their families to make psychological preparations and subsequent treatment plans. Moreover, the expression level of Ki-67 aids physicians in assessing the proliferation rate and malignancy of gliomas, thereby facilitating the formulation of more appropriate surgical plans. For gliomas with high Ki-67 LI, a more extensive surgical resection may be necessary to reduce the risk of tumor recurrence (9). These pieces of information obtained through preoperative non-invasive magnetic resonance imaging (MRI) assessment are irreplaceable by those derived from postoperative pathology.

So accurate assessment of Ki-67 LI is instrumental in the diagnosis and treatment of glioma. A study reported that Ki-67 LI greater than 5% are associated with malignant progression and poor prognosis, it has been revealed that Ki-67 LI is an independent prognostic factor for survival (6). Hence, classifying the Ki-67 LI status according to 5% would be practical and meaningful.

However, currently the “gold standards” for identifying Ki-67 LI is immunohistochemical analysis, which is invasive. Another crucial point is that glioma is a tumor with extremely strong heterogeneity; the accuracy of tumor pathology results, the final synthesis of prognostic information, depends critically on accurate biopsy sampling of the highest grade of disease present, and is vulnerable to undersampling (10). Targeting highly proliferative areas should increase the likelihood of sampling highly malignant tumor areas, which will improve the accuracy of prognosis and treatment (11).

The proliferation level of Ki-67 serves as one of the reflective forms of cellular and tumor microstructural composition, and diffusion MRI happens to be currently one of the most powerful magnetic resonance technique capable of observing the diffusion of water molecules within biological tissues and assessing the integrity of cellular membranes in vivo (12).

Diffusion tensor imaging (DTI) is one of the earliest and well-established diffusion imaging techniques, capable of reflecting cellular density to a certain extent (13). And it has been widely used in clinical and research settings to assess white matter fibers (12). It is a signal representation valid at low b values (b≤1,000 s/mm²), derived from the second-order term of the cumulant expansion. It does not assume Gaussian diffusion but reflects the dominant diffusion anisotropy at the voxel scale (14). The motion of water molecules within biological organisms, given sufficient time, follows a non-Gaussian distribution (15), which gave rise to the emergence of diffusion kurtosis imaging (DKI).

DKI posits that water molecules in neural tissue move in a non-Gaussian (restricted and hindered) pattern (16). DKI more accurately reflects the true diffusion of water molecules in biological tissues and better captures the complexity and heterogeneity of the tissue microenvironment. DKI reflects the microstructural tissue complexity (12,16). DTI and DKI metrics characterize water diffusion in both gray and white matter, though their interpretation varies with tissue architecture. For example, fractional anisotropy (FA) in white matter reflects axonal coherence, while in gray matter, it may arise from dendritic or glial organization.

Neurite orientation dispersion and density imaging (NODDI) (17) model is a novel diffusion imaging technique developed in recent years based on models of hindered and restricted diffusion of water molecules (2,18-20). The distinction between NODDI and DKI/DTI lies in their theoretical foundations: DTI/DKI are signal representations agnostic to microstructure, whereas NODDI is a biophysical model assuming three compartments (intra-neurite, extra-neurite, and free water) with predefined diffusivities and orientation distributions (14). It enables the investigation of gray matter microstructure by modeling water diffusion into three compartments: intra-neurite, extra-neurite, and cerebrospinal fluid, thereby describing the microstructure of white and gray matter and reflecting the complexity of gray matter regions and nuclei (18).

Due to the highly heterogeneous nature of gliomas, the integration of these three models, based on their respective principles, can comprehensively evaluate the microstructural characteristics of tissues, encompassing both gray matter and white matter.

Many previous studies have shown that quantitative MRI metrics [such as apparent diffusion coefficient (ADC) value (21,22)] could potentially be used as prognostic biomarkers in glioma (23-26). Diffusion-weighted imaging (DWI) has emerged as a pivotal tool for non-invasive evaluation of tumor proliferation, with the ADC demonstrating a consistent inverse correlation with Ki-67 labeling index (LI) across gliomas (9). Lower ADC values, particularly ADCmin, reflect restricted water diffusion in hypercellular, high Ki-67 gliomas (27). Beyond DWI, advanced diffusion models such as DTI (28) and DKI have further refined this relationship: FA (29) and mean kurtosis (Kmean) (30-32) correlate positively with Ki-67 LI (r=0.76 for Kmean) (33), capturing microstructural complexity and tumor invasiveness (34). Recent innovations, including multi-shell techniques like propagator-based MAP-MRI (28,31), have expanded the biomarker repertoire—parameters such as return-to-origin probability (RTOP) show promise in quantifying proliferative heterogeneity. Ke et al. (35) combined ADC with texture features; they improved Ki-67 prediction values with area under the receiver operating characteristic curves (AUC) of 0.92. Delta radiomics from dynamic contrast-enhanced MRI (DCE-MRI) combined with ADC enhances Ki-67 LI accuracy (AUC =0.800) (9).

Although some prognostic biomarkers or prediction models have been developed in gliomas, to our best knowledge, there are few studies on predicting Ki-67 LI by combining multimodal diffusion models, especially utilizing histogram method.

Our research innovatively combined multi-modal diffusion MRI to comprehensively investigate the diagnostic efficiency for predicting Ki-67 LI in adult diffuse gliomas. By integrating multiple diffusion MRI techniques, we aim to improve the accuracy and reliability of Ki-67 LI predictions, thereby advancing our understanding of glioma biology and facilitating personalized treatment strategies. As there are few prior studies that have investigated multi-modal diffusion MRI for Ki-67 LI predicting in glioma patients, the present study can be considered a pilot study for the feasibility of this method. On the other hand, the spatial heterogeneity of diffuse gliomas poses significant challenges to accurate Ki-67 assessment, as histopathological sampling may overlook regions of high proliferation. To address this, we employ histogram analysis of diffusion MRI metrics, which quantifies intra-tumoral heterogeneity by capturing the voxel-wise distribution of parameters across the entire lesion. This approach aims to provide a more representative estimate of global proliferative activity.

Hence, the purpose of our study was to develop a nomogram model combining three mainstream diffusion MRI variables to predict Ki-67 LI, and further evaluate the predictive value of this model for the overall survival (OS) prognosis of glioma patients. We present this article in accordance with the TRIPOD+AI reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-242/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the institutional review board of The First Affiliated Hospital of Sun Yat-sen University (No. [2021]209) and individual consent for this retrospective analysis was waived. And this single-center retrospective study was registered in the ClinicalTrials.gov database (Identifier: NCT06572592) prior to participant enrollment.

Study cohorts

From our institute of a tertiary referral university hospital, consecutive adult patients with diffuse glioma who underwent conventional MRI and whole-brain diffusion examination within three weeks before surgery between May 2014 and May 2021 were considered for inclusion. The follow-up endpoint for patients is May 2023. Median follow-up from inclusion was 21.1 months [95% confidence interval (CI): 16.48–24.15]. Median OS from inclusion was 20.69 months (95% CI: 13.93–24.50).

The following exclusion criteria were rigorously applied to ensure the accuracy and reliability of our data:

• Participants who had undergone any therapy, such as surgery, chemotherapy, steroid treatment, or biopsy, before MRI examination were excluded, as these treatments could potentially alter the imaging features of the tumor and bias our results.
• Participants under 18 years of age were excluded, as our study focused on adult patients with diffuse glioma.
• MR images with noticeable artifacts were excluded, as these artifacts could interfere with the accuracy of diffusion MRI metrics and affect the reliability of our analysis.
• Tumors with a volume less than 20 mm³ were excluded, as smaller tumors may not provide sufficient data points for accurate histogram analysis.

Clinical-pathologic data collection

Clinical data on age and gender were collected. Pathologic data on Ki-67 LI, grade were collected from reports of biopsies performed after resection. Simple imaging features such as the maximum diameter of the tumor, whether the ependyma was invaded, and whether there was contralateral brain invasion were recorded through T1WI, T2WI, T2WI fluid-attenuated inversion recovery (FLAIR), and T1WI enhanced images.

Image acquisition and analysis

Participants underwent whole-brain DWI and conventional MRI examination with the 3.0-T scanner (Magnetom Prisma and Magnetom Verio, Siemens Healthcare) with a 64-channel (Magnetom Prisma) and 12-channel head-neck coil (Magnetom Verio), respectively. The specific values of the scanning parameters are provided in Appendix 1. All MRI scans were conducted under the supervision of dedicated radiologists who specialized in the field. The radiologists not only tracked the scanning process but also conducted rigorous image quality assessments during and after scans. The quality assessments included evaluations of key parameters such as signal-to-noise ratio, contrast-to-noise ratio, spatial resolution, and the presence of artifacts or distortions. All cases included in our study were of high quality and suitable for accurate quantitative analysis.

Image analysis was retrospectively performed in a blinded manner without clinical-pathologic information being made available. An image review was performed by two board-certified radiologists in consensus (with 6 and 20 years of experience, respectively). A third board-certified radiologist (with 3 years of experience) independently delineated the regions of interest (ROIs) for the purpose of intraclass correlation coefficient (ICC) assessment.

All the imaging data formats were converted to NIFTI format and the diffusion-weighted images were co-registered to the nondiffusion-weighted (b=0) images to minimize the artifacts by eddy current correction (36,37). NODDI parameter diagrams [including Ficvf (intracellular volume fraction), Odi (orientation dispersion) and Fiso (volume fraction of the isotropic compartment)] were exported by using corresponding NODDI’s Matlab toolbox (http://www.nitrc.org/projects/noddi_toolbox). DTI and DKI fitting were performed using diffusion-weighted images with different b values. DKE software (http://www.nitrc.org/projects/dke, version 2.5.1) was used to calculate DKI parameter diagrams [including Kmean, Kr (radial kurtosis) and Ka (axial kurtosis)] and DTI parameter diagrams [including mean diffusivity (MD) and FA]. Each parameter image was registered with axial T2 FLAIR image or 3D-T1 MPRANGE (magnetization prepared rapid gradient echo) image by voxel-based nonlinear registration method. Histogram analysis was performed by using software ImageJ (Version 1.46r, NIH, USA). The ROIs were manually drawn on tumor parenchyma on the slice of maximum tumor size by consensus between two experienced neuroradiologists (blinded to all pathological and clinical outcomes), with 6 years of experience in neuroradiology and 20 years of experience, respectively. For lesions with enhancement, we performed ROI delineation on the co-registered T1WI-CE images. For lesions without enhancement, we delineated the ROIs on the co-registered T2WI FLAIR images. Large vessels, meninges, necrotic cystic lesions and bleeding areas were avoided, as shown in Figure 1. The ROIs were automatically transferred to the corresponding diffusion parameters.

Figure 1 An example of ROI placement. A 32-year-old man with pathologically diagnosed diffuse astrocytoma (WHO grade 2). According to the co-registration transversal T2-FLAIR, the ROI in tumor parenchyma was outlined at the maximum successive level. Large vessels, meninges, necrotic cystic lesions and bleeding areas were avoided. The ROI was copied simultaneously to the registered DTI (FA, MD), DKI (Ka, Kmean, Kr) and NODDI (Ficvf, Odi, Fiso) maps. DKI, diffusion kurtosis imaging; DTI, diffusion tensor imaging; FA, fractional anisotropy; Ficvf, intracellular volume fraction; Fiso, volume fraction of the isotropic compartment; FLAIR, fluid-attenuated inversion recovery; Ka, axial kurtosis; Kmean, mean kurtosis; Kr, radial kurtosis; MD, mean diffusivity; NODDI, neurite orientation dispersion and density imaging; Odi, orientation dispersion; ROI, region of interest; WHO, World Health Organization.

Histogram analysis was performed using ImageJ software (Version 1.46r). A total of eight diffusion metrics (FA and MD from DTI model; Kmean, Ka, and Kr from DKI model; Ficvf, Odi, and Fiso from NODDI model) consisting of the mean value and standard deviation were calculated from ROIs overlaid on corresponding diffusion metric maps. All the diffusion histograms were then generated for the matched ROIs. All the diffusion histograms were then generated for the matched ROIs. The analysis included a mean of 8,388.76±5,957.83 pixels per patient (range, 630–28,380 pixels). The histograms were normalized by total pixel count to account for inter-patient ROI size variations, and the number of bins was standardized to 256 based on the Image Biomarker Standardization Initiative (IBSI) guidelines (38,39). The following histogram parameters were extracted: (I) 10^th, 20^th, 25^th, 75^th, 80^th and 90^th centile points (P10, P20, P25, P75, P80 and P90), where the X^th percentile where the Xth percentile represents the value below which X% of the voxel values in the histogram are located (39); (II) skewness and kurtosis.

Immunohistochemical studies

Ki-67 LI was determined by utilizing the immunohistochemistry Envision method (Clone No. UMAB107, dilution 1:300). The tumor sections were quantified based on the percentage of positive cells (cells with nuclear staining intensity) in the highest density staining area. The Ki-67 LI was defined as the percentage of positive cells among the total cells counted (40). According to previous studies (41,42), Ki-67 LI was classified into low: ≤5% and high: >5% (6).

Statistical analysis

The dataset was stratified by Ki-67 LI status (high/low) and randomly divided into a training set and a test set at a ratio of 7:3, preserving the original proportion of Ki-67 LI subgroups in both sets. The data in the training set were standardized using mean normalization and variance scaling, and subsequently, the data in the test set were standardized using the same normalization method as applied to the training set. ICC was calculated to evaluate interobserver reproducibility. Histogram features with good interobserver reproducibility (ICC >0.75) were selected for comparison between different Ki-67 LI groups. Clinical features (age and gender) and imaging features (histogram features of three diffusion model metrics as well as simple imaging features) were compared between the two groups (Ki-67 LI high group and Ki-67 LI low group) using the Wilcoxon rank-sum test or the independent t-test in the training set. All univariate comparisons were adjusted using the Benjamini-Hochberg false discovery rate (FDR) procedure (q<0.05). Further, feature selection in the training set was conducted using the recursive feature elimination (RFE) method, with logistic regression as the model and 5-fold cross-validation to achieve the optimal accuracy value, which were incorporated to construct a nomogram model. The performance of the multivariable nomogram model was assessed for discrimination in the training set and test set, respectively. Discrimination was quantified by using the area under the receiver operating characteristic curve (AUC).

Gender, tumor size, contralateral parenchymal invasion status, ventricular ependymal invasion status, IDH mutation status, extent of resection, adjuvant radiotherapy and chemotherapy were enrolled into the univariable Cox regression analysis. Statistical tests were two-sided, P<0.05 was statistically significant.

In the outcome cohort, Youden’s index method was utilized to identify the cut-off value (1.314) of the risk score, aiming to achieve the maximum accuracy value for predicting OS, thus classifying patients into high-risk and low-risk groups. Cox regression was used to generate survival curves in the training set and the test set, respectively. Statistical analyses were performed using open-source software Python and R (Version 4.3.2). A two-sided P value less than 0.05 was indicative of a significant difference.

Results

Patient characteristics

A total of 156 patients who underwent surgical resection were eligible for this study; 33 were excluded owing to receiving other treatment before surgery (n=21), poor-quality of radiological images (n=7) or measurable target lesion of less than 20 mm³ (n=5), as shown in Figure 2. Accordingly, 123 patients (39 women; mean age, 46.08±13.24 years) were included. The clinicopathological characteristics of patients in the training set (n=86) and the test set (n=37) are summarized in Table 1. The two data sets were similar in the distribution of characteristics. Patients in the high Ki-67 LI group tend to have a higher average age compared to those in the low Ki-67 LI group (all data set P<0.05).

Figure 2 Flowchart for the patient recruitment process. MRI, magnetic resonance imaging.

In the outcome cohorts, all these 123 patients were included. The baseline characteristics in the outcome cohort are listed in Table 1. The median follow-up from inclusion was 21.1 months (95% CI: 16.48–24.15). Median OS from inclusion were 20.69 months (95% CI: 13.93–24.50).

Construction of the diffusion MRI nomogram model

Comparison between groups with different Ki-67 LI levels in the training set revealed significant differences in nearly all diffusion histogram features (P<0.05). After RFE analysis, eight features (Fiso Skewness, Fiso P90, Kmean P75, Kmean P10, Kmean Kurtosis, Odi P75, Ka P80 and Ficvf Kurtosis) were obtained and were used to develop a nomogram model. And the intergroup comparisons of eight parameters incorporated into the nomogram model are shown in Table 2. The simple imaging features, patient age, gender, and other clinical characteristics were ranked lower in the order of feature importance and were not included in the final nomogram model construction. Figure 3 shows an example of histogram distributions of DKI/NODDI metrics for two representative cases: Patient A: 42-year-old male with Ki-67 LI =8% (relatively high proliferation cohort); Patient B: 32-year-old female with Ki-67 LI =3% (relatively low proliferation cohort). Relatively high proliferation group’s (Patient A as an example in Figure 3) histogram is shifted to the right for Ka, Kmean, Kr, and Ficvf values, indicating a higher distribution. The broader base and lower peak frequency in relatively high proliferation group (Patient A as an example in Figure 3) also indicated stronger heterogeneity.

Figure 3 An example of histogram distributions of DKI/NODDI metrics for two representative cases. Patient A: 42-year-old male with Ki-67 LI =8% (relatively high proliferation cohort). Patient B: 32-year-old female with Ki-67 LI =3% (relatively low proliferation cohort). Patient A’s histogram is shifted to the right for Ka, Kmean, Kr, and Ficvf values, indicating a higher distribution. The broader base and lower peak frequency in Patient A also indicated stronger heterogeneity. DKI, diffusion kurtosis imaging; FA, fractional anisotropy; Ficvf, intracellular volume fraction; Fiso, volume fraction of the isotropic compartment; Ka, axial kurtosis; Kmean, mean kurtosis; Kr, radial kurtosis; LI, labelling index; MD, mean diffusivity; NODDI, neurite orientation dispersion and density imaging; Odi, orientation dispersion.

The distribution of the coefficient in each data set and feature importance selection results are given in Figure 4A. The nomogram model that integrated the eight diffusion histogram variables was developed. The receiver operating characteristic curves of the models for predicting the Ki-67 LI groups showed stable performance in both training set and the test set (training set: AUC=0.92, 95% CI: 0.85–0.98, sensitivity =0.85, specificity =0.84; test set: AUC=0.84, 95% CI: 0.64–0.98, sensitivity =0.77, specificity =0.73) (Figure 4B,4C).

Figure 4 Diffusion MRI histogram model for Ki-67 LI prediction: feature selection, ROC validation, and clinical nomogram. (A) Optimal features after RFE analysis. (B) Performance of the nomogram model for predicting Ki-67 LI with ROC curve analysis in the training set and test set. (C) The diffusion MRI based histogram model presented with a nomogram scaled by the proportional regression coefficient of each predictor. AUC, area under the curve; Ficvf, intracellular volume fraction; Fiso, volume fraction of the isotropic compartment; Ka, axial kurtosis; Kmean, mean kurtosis; LI, labelling index; MRI, magnetic resonance imaging; Odi, orientation dispersion; RFE, recursive feature elimination; ROC, receiver operating characteristic.

The nomogram model score was lower for the Ki-67 LI low group than for the high group in the training set (0.57±0.61 vs. 1.73±0.74; P<0.001), and the test set (0.66±0.72 vs. 1.71±0.71; P<0.001) (Figure 5).

Figure 5 Nomogram model score comparisons in training and test sets. Low vs. high Ki-67 LI groups in the (A) training set and (B) test set. (C) LGG vs. HGG in the (C) training set and (D) test set. HGG, high grade glioma; LI, labelling index; LGG, low grade glioma.

Secondary observation about glioma grade

Furthermore, the nomogram model score also showed good discrimination in glioma grade. The nomogram model score was lower for low-grade gliomas (LGGs) than for high-grade gliomas (HGGs) in the training set (0.57±0.32 vs. 0.87±0.13; P<0.001), and the test set (0.61±0.33 vs. 0.85±0.15; P<0.01) (Figure 5).

Predictive value of the nomogram model for outcomes

The median follow-up period was 21.1 months (95% CI: 16.48–24.15). Median OS from inclusion was 20.69 months (95% CI: 13.93–24.50).

Univariable Cox regression analysis demonstrated that the IDH mutation status was significantly associated with survival time [hazard ratio (HR) 0.34, 95% CI: 0.18–0.64, P<0.001]. However, gender, tumor size, contralateral parenchymal invasion, ventricular ependymal invasion, extent of resection, radiotherapy and chemotherapy showed no statistical significance in OS (P≥0.05) (Table 3),

According to cut-off value (1.31) of the risk score, patients in the training set were classified into high-risk and low-risk groups. The baseline nomogram model score was higher in patients in the high-risk group than those in the low-risk group (0.53±0.36 vs. 2.11±0.40; P=0.001) (Figure 6A). Patients in the low-risk group showed a longer OS than those in the high-risk group (HR: 5.36, P<0.001) (Figure 6B), which was confirmed in the test set. Kaplan-Meier analysis showed a statistically significant difference in OS between low and high nomogram model scores (P<0.001), both in the training set and the test set (Figure 6C,6D).

Figure 6 Performance of the nomogram model for risk score in the outcome cohort. The difference of nomogram model score between the low risk score group (group 0) and high risk score group (group 1) in the training set (A) and the test set (B). Cox regression analysis of overall survival according to dichotomized nomogram model score (low or high) in the training set (C) and the test set (D). ***, P<0.001.

Discussion

Our study revealed that quantitative histogram analysis of multi-modal diffusion MRI could effectively predict the Ki-67 LI in adult-type diffuse gliomas. We constructed a nomogram model to noninvasively predict Ki-67 LI with areas under the receiver operating characteristic curve of 0.92 (95% CI: 0.85–0.98) and 0.84 (95% CI: 0.64–0.98) in the training set and the test set, respectively. The nomogram model we obtained can calculate a nomogram score for each patient. We further applied this model to glioma grading, and we found that there were substantial differences in nomogram scores between HGG and LGG. More importantly, we also showed that a low baseline nomogram model score was associated with longer OS in glioma patients, indicating its potential clinical application.

The level of Ki-67 expression is closely correlated with the recurrence risk of gliomas. Patients with high Ki-67 LI indices generally have poorer prognoses, lower cure rates, and shorter survival times (43). Additionally, the Ki-67 LI can be used as a reference for developing treatment plans. Patients with high Ki-67 LI may require more intensive treatment strategies and more aggressive therapeutic measures (44). The level of Ki-67 expression aids clinicians in assessing the proliferative capacity and malignant potential of gliomas, thereby enabling the formulation of more precise surgical plans. For gliomas exhibiting high Ki-67 LI expression, a more extensive surgical resection may be indicated to minimize the risk of tumor recurrence. However, the current evaluation of Ki-67 LI mainly relies on pathological assessment (45), which is invasive. Additionally, gliomas are highly heterogeneous tumors, thus the local pathological tissue cannot reflect the overall condition. Therefore, it is of particular importance to assess Ki-67 LI through non-invasive and comprehensive methods.

Preoperative MRI examination has been recognized as an essential digital biopsy approach to predict the biological features of tumors (4,46,47). DWI, one of the most significant functional magnetic resonance imaging techniques, has been widely used for brain tumor evaluation. Currently, several mainstream models, including DTI, DKI and NODDI, are utilized to provide comprehensive analysis of lesions from multiple dimensions (17,48,49). Importantly, these models enable quantitative analysis. However, to our knowledge, there were few studies integrating these three diffusion models to predict Ki-67 LI.

Based on previous research, a potential correlation between quantitative DKI and NODDI parameters and Ki-67 LI has been revealed (17). We have further explored this relationship by combining three diffusion models and extracting histogram features to construct a nomogram model for predicting Ki-67 LI. After applying RFE method for feature selection, 8 quantitative parameters were selected from a total of 80 histogram parameters in the diffusion models to construct our final model. And this final model included 4 parameters from DKI (Kmean P75, Kmean P10, Kmean Kurtosis, and Ka P80) and 4 parameters from NODDI (Fiso Skewness, Fiso P90, Odi P75, and Ficvf Kurtosis) model, which showed good performance with AUC of 0.92 and 0.84 in the training set and test set, respectively.

Kmean is defined as the average of diffusion kurtosis values along various directions within the tissue (50) and Ka specifically referring to the kurtosis value associated with diffusion along the axial direction (50). The increase in Kmean and Ka values could be explained by the high Ki-67 LI, which is typically associated with increased intracellular cell density, nuclear pleomorphism, and microvascular proliferation (51). These factors restrict water molecule movement and contribute to elevated Kmean and Ka values.

Fiso represents the proportion of the isotropic component in the diffusion signal, which is associated with diffusion in extracellular spaces such as cerebrospinal fluid or edematous tissue (50). As Ki-67 LI increases, severe tissue edema may develop, increasing membrane permeability (51) and facilitating the accumulation of free water. This process elevates the extracellular fluid content, thereby increasing Fiso values. Furthermore, higher Ki-67 LI values also suggest more extensive tissue necrosis (51) and severe neurite injury, reducing diffusion restriction within neurites, which further contributes to the increase in Fiso values.

Ficvf represents the intracellular volume fraction, indicating the proportion of intracellular (neurite-related) volume relative to the total volume (50). With increasing Ki-67 LI, heightened tumor cell proliferation compresses the extracellular space (51). As a result, neurites are packed more densely, increasing cell density and subsequently elevating Ficvf values.

ODI stands for the orientation dispersion index, reflecting the degree of neurite orientation heterogeneity (50). As Ki-67 LI increases, the severity of nerve fiber injury intensifies, leading to greater dispersion in neurite orientations and increased heterogeneity in the tumor tissue (51). This reduced directional coherence is reflected in elevated ODI values.

Notably, no parameters from the DTI model were selected in the nomogram model after RFE feature selection, as well as simple imaging features, patient age, gender, or other clinical characteristics. This finding suggests that higher-order diffusion models may provide a more comprehensive reflection of the pathological characteristics of tumors.

Based on our nomogram model, each patient is assigned a nomogram score. When this nomogram score obtained from our model is further applied to tumor grading, we observe significant differences in the nomogram scores between HGGs and LGGs. This finding further expands the clinical application capabilities of our model.

Further nomogram analysis showed that a low nomogram model score, likely indicative of low Ki-67 LI, was associated with longer OS in adult-type diffuse glioma patients according to the cut-off value of 1.31 of the risk score. This enables us to calculate a corresponding nomogram risk score for each glioma patient after preoperative diffusion magnetic resonance imaging (MRI) based on their MRI results. Using a cut-off value of 1.31 for this score, we can predict the patient’s long or short survival duration. And the results were reproducible, indicating that the approach may be generalizable to other patient samples.

Our research extends the current understanding by establishing a correlation between diffusion MRI, Ki-67 LI, and clinical outcomes. This correlation holds promise for integrating morphomolecular subtyping into clinical routines and possesses the potential to aid in the prediction of patients who shows better prognosis. By integrating histogram features from multiple diffusion models, our nomogram statistically aggregates heterogeneous subregional properties, thereby reducing the sampling bias inherent to focal biopsy. This is particularly relevant in gliomas, where discordance between imaging and histopathology often reflects tumor heterogeneity rather than methodological error. In addition, exploring the possibility of combining our model with other biomarkers is another promising avenue for future research. By integrating additional biomarkers that reflect different aspects of the disease process, we could potentially enhance the predictive power of our model. For instance, combining genetic (such as IDH1/2 and methylation of the MGMT promoter. Patients harboring IDH1/2 mutations exhibit relative sensitivity to radiotherapy and alkylating agents, and also represent potential therapeutic targets for Ivosidenib (52). Methylation of the MGMT promoter is associated with a favorable prognosis in glioblastoma and enhanced responsiveness to temozolomide treatment (53). Additionally, proteomic analyses, such as evaluating programmed death-ligand 1 (PD-L1) expression to predict patient response to programmed cell death-1 (PD-1)/PD-L1 inhibitor therapy (54), could provide a more comprehensive view of the disease and improve our ability to stratify patients accurately. Furthermore, such an integrated approach could lead to the development of more personalized treatment strategies, tailored to the individual needs of patients.

While histogram parameters (e.g., skewness, kurtosis) reflect global tumor heterogeneity, they inherently lack spatial localization information. However, these features can indeed serve as a foundation for generating segmented maps that visualize spatial variations in tumor biology. Existing studies demonstrate that histogram-derived metrics can guide segmentation algorithms to highlight ROIs with distinct biological properties, and spatial maps derived from histogram features could optimize biopsy targeting by identifying regions with elevated proliferation potential. For example, Kim et al. (55) integrated texture and wavelet features with histogram analysis in compressed domains to improve segmentation accuracy in colorectal cancer biopsies, emphasizing the value of combining histogram features with spatial filters. We fully agree that extending histogram analysis to spatial mapping holds significant clinical potential. Future efforts will explore this direction, leveraging advanced segmentation and validation frameworks to enhance preoperative planning for glioma patients.

There are several limitations in this study. First, the proportion of sample sizes in the two Ki-67 LI groups was inconsistent (Ki-67 LI group high: 87 and group low: 36). The sample size was relatively small. A larger sample size is needed to verify the results. Second, the cohort was constructed based on retrospectively gathered data from a single institution, which may introduce selection bias; thus, further multicenter validation research is required. Third, the 2D slice-based ROIs used in this study may not fully capture the volumetric complexity of gliomas. However, manual 2D ROI delineation remains a widely adopted practice in glioma imaging studies due to its efficiency and reproducibility, particularly when guided by multimodal diffusion MRI to target regions of high cellularity or perfusion heterogeneity. Furthermore, our histogram parameters (e.g., skewness, kurtosis) inherently aggregate voxel-level heterogeneity across the ROI, mitigating spatial sampling bias to some extent. Future work should adopt automated 3D segmentation or radiomics to improve tumor-wide characterization and clinical applicability.

Conclusions

Our research innovatively combined multi-modal diffusion MRI to comprehensively investigate the diagnostic efficiency for predicting Ki-67 LI in adult diffuse gliomas. In conclusion, the nomogram model based on quantitative histogram analysis of three diffusion MRI model (DTI, DKI and NODDI) could noninvasively predict Ki-67 LI and indicate clinical outcomes in patients with glioma, which may serve as imaging biomarkers. This prediction model has great potential to guide clinical prognosis prediction and decision-making for therapy in the future.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD+AI reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-242/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-242/dss

Funding: This study received funding by National Natural Science Foundation of China (NSFC 82172015, No 82202217), Natural Science Foundation of Guangdong Province (No. 2022A1515011264), Fundamental Research Funds for the Central Universities, Sun Yat-sen University (No. 23xkjc024), Beijing Xisike Clinical Oncology Research Foundation (No. Y-zai2022/qn-0250) and Guangdong Basic and Applied Basic Research Foundation (No. 2024A1515013041).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-242/coif). The 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 study was approved by the institutional review board of The First Affiliated Hospital of Sun Yat-sen University (No. [2021]209) and individual consent for this retrospective analysis was waived.

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. Louis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, Hawkins C, Ng HK, Pfister SM, Reifenberger G, Soffietti R, von Deimling A, Ellison DW. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro Oncol 2021;23:1231-51. [Crossref] [PubMed]
2. Li G, Li L, Li Y, Qian Z, Wu F, He Y, Jiang H, Li R, Wang D, Zhai Y, Wang Z, Jiang T, Zhang J, Zhang W. An MRI radiomics approach to predict survival and tumour-infiltrating macrophages in gliomas. Brain 2022;145:1151-61.
3. Neftel C, Laffy J, Filbin MG, Hara T, Shore ME, Rahme GJ, et al. An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell 2019;178:835-849.e21. [Crossref] [PubMed]
4. Weller M, van den Bent M, Preusser M, Le Rhun E, Tonn JC, Minniti G, et al. EANO guidelines on the diagnosis and treatment of diffuse gliomas of adulthood. Nat Rev Clin Oncol 2021;18:170-86. [Crossref] [PubMed]
5. Habberstad AH, Gulati S, Torp SH. Evaluation of the proliferation markers Ki-67/MIB-1, mitosin, survivin, pHH3, and DNA topoisomerase IIα in human anaplastic astrocytomas--an immunohistochemical study. Diagn Pathol 2011;6:43. [Crossref] [PubMed]
6. Jaros E, Perry RH, Adam L, Kelly PJ, Crawford PJ, Kalbag RM, Mendelow AD, Sengupta RP, Pearson AD. Prognostic implications of p53 protein, epidermal growth factor receptor, and Ki-67 labelling in brain tumours. Br J Cancer 1992;66:373-85. [Crossref] [PubMed]
7. Yang X, Hu C, Xing Z, Lin Y, Su Y, Wang X, Cao D. Prediction of Ki-67 labeling index, ATRX mutation, and MGMT promoter methylation status in IDH-mutant astrocytoma by morphological MRI, SWI, DWI, and DSC-PWI. Eur Radiol 2023;33:7003-14. [Crossref] [PubMed]
8. Grabner G, Kiesel B, Wöhrer A, Millesi M, Wurzer A, Göd S, Mallouhi A, Knosp E, Marosi C, Trattnig S, Wolfsberger S, Preusser M, Widhalm G. Local image variance of 7 Tesla SWI is a new technique for preoperative characterization of diffusely infiltrating gliomas: correlation with tumour grade and IDH1 mutational status. Eur Radiol 2017;27:1556-67. [Crossref] [PubMed]
9. Du N, Shu W, Li K, Deng Y, Xu X, Ye Y, Tang F, Mao R, Lin G, Li S, Fang X. An initial study on the predictive value using multiple MRI characteristics for Ki-67 labeling index in glioma. J Transl Med 2023;21:119. [Crossref] [PubMed]
10. Jackson RJ, Fuller GN, Abi-Said D, Lang FF, Gokaslan ZL, Shi WM, Wildrick DM, Sawaya R. Limitations of stereotactic biopsy in the initial management of gliomas. Neuro Oncol 2001;3:193-200. [Crossref] [PubMed]
11. Gates EDH, Lin JS, Weinberg JS, Hamilton J, Prabhu SS, Hazle JD, Fuller GN, Baladandayuthapani V, Fuentes D, Schellingerhout D. Guiding the first biopsy in glioma patients using estimated Ki-67 maps derived from MRI: conventional versus advanced imaging. Neuro Oncol 2019;21:527-36. [Crossref] [PubMed]
12. Hoffmann E, Masthoff M, Kunz WG, Seidensticker M, Bobe S, Gerwing M, Berdel WE, Schliemann C, Faber C, Wildgruber M. Multiparametric MRI for characterization of the tumour microenvironment. Nat Rev Clin Oncol 2024;21:428-48. [Crossref] [PubMed]
13. Basser PJ. Inferring microstructural features and the physiological state of tissues from diffusion-weighted images. NMR Biomed 1995;8:333-44. [Crossref] [PubMed]
14. Novikov DS, Kiselev VG, Jespersen SN. On modeling. Magn Reson Med 2018;79:3172-93. [Crossref] [PubMed]
15. Zhuo J, Xu S, Proctor JL, Mullins RJ, Simon JZ, Fiskum G, Gullapalli RP. Diffusion kurtosis as an in vivo imaging marker for reactive astrogliosis in traumatic brain injury. Neuroimage 2012;59:467-77. [Crossref] [PubMed]
16. Hui ES, Cheung MM, Qi L, Wu EX. Towards better MR characterization of neural tissues using directional diffusion kurtosis analysis. Neuroimage 2008;42:122-34. [Crossref] [PubMed]
17. Zhao J, Li JB, Wang JY, Wang YL, Liu DW, Li XB, Song YK, Tian YS, Yan X, Li ZH, He SF, Huang XL, Jiang L, Yang ZY, Chu JP. Quantitative analysis of neurite orientation dispersion and density imaging in grading gliomas and detecting IDH-1 gene mutation status. Neuroimage Clin 2018;19:174-81. [Crossref] [PubMed]
18. Zhang H, Schneider T, Wheeler-Kingshott CA, Alexander DC. NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain. Neuroimage 2012;61:1000-16. [Crossref] [PubMed]
19. Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, Ma ZL, Liu ZY. Development and Validation of a Radiomics Nomogram for Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer. J Clin Oncol 2016;34:2157-64. [Crossref] [PubMed]
20. Liu X, Li Y, Qian Z, Sun Z, Xu K, Wang K, Liu S, Fan X, Li S, Zhang Z, Jiang T, Wang Y. A radiomic signature as a non-invasive predictor of progression-free survival in patients with lower-grade gliomas. Neuroimage Clin 2018;20:1070-7. [Crossref] [PubMed]
21. Wang X, Shu X, He P, Cai Y, Geng Y, Hu X, Sun Y, Xiao H, Zheng W, Song Y, Xue Y, Jiang R. Ultra-high b-value DWI accurately identifies isocitrate dehydrogenase genotypes and tumor subtypes of adult-type diffuse gliomas. Eur Radiol 2024;34:6751-62. [Crossref] [PubMed]
22. She Y, Liu X, Jiang J, Wang X, Niu Q, Zhou J. The role of apparent diffusion coefficient in the grading of adult isocitrate dehydrogenase-mutant astrocytomas: relationship with the Ki-67 proliferation index. Acta Radiol 2024;65:489-98. [Crossref] [PubMed]
23. El-Habashy DM, Wahid KA, He R, McDonald B, Mulder SJ, Ding Y, Salzillo T, Lai SY, Christodouleas J, Dresner A, Wang J, Naser MA, Fuller CD, Mohamed ASRJoint Head and Neck Radiation Therapy-MRI Development Cooperative. Dataset of weekly intra-treatment diffusion weighted imaging in head and neck cancer patients treated with MR-Linac. Sci Data 2024;11:487. [Crossref] [PubMed]
24. Fang S, Yang Y, Tao J, Yin Z, Liu Y, Duan Z, Liu W, Wang S. Intratumoral Heterogeneity of Fibrosarcoma Xenograft Models: Whole-Tumor Histogram Analysis of DWI and IVIM. Acad Radiol 2023;30:2299-308. [Crossref] [PubMed]
25. Lafcı O, Celepli P, Seher Öztekin P, Koşar PN. DCE-MRI Radiomics Analysis in Differentiating Luminal A and Luminal B Breast Cancer Molecular Subtypes. Acad Radiol 2023;30:22-9. [Crossref] [PubMed]
26. Zhang L, Liu X, Xu X, Liu W, Jia Y, Chen W, Fu X, Li Q, Sun X, Zhang Y, Shu S, Zhang X, Xiang R, Chen H, Sun P, Geng D, Yu Z, Liu J, Wang J. An integrative non-invasive malignant brain tumors classification and Ki-67 labeling index prediction pipeline with radiomics approach. Eur J Radiol 2023;158:110639. [Crossref] [PubMed]
27. Zheng Y, Huang W, Zhang X, Lu C, Fu C, Li S, Lin G. A Noninvasive Assessment of Tumor Proliferation in Lung cancer Patients using Intravoxel Incoherent Motion Magnetic Resonance Imaging. J Cancer 2021;12:190-7. [Crossref] [PubMed]
28. Sun Y, Su C, Deng K, Hu X, Xue Y, Jiang R. Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status. Eur Radiol 2022;32:3744-54. [Crossref] [PubMed]
29. Zhu H, Xie Y, Li L, Liu Y, Li S, Shen N, Zhang J, Yan S, Liu D, Li Y, Zhu W. Diffusion along the perivascular space as a potential biomarker for glioma grading and isocitrate dehydrogenase 1 mutation status prediction. Quant Imaging Med Surg 2023;13:8259-73. [Crossref] [PubMed]
30. Li S, Zheng Y, Sun W, Lasič S, Szczepankiewicz F, Wei Q, Han S, Zhang S, Zhong X, Wang L, Li H, Cai Y, Xu D, Li Z, He Q, van Westen D, Bryskhe K, Topgaard D, Xu H. Glioma grading, molecular feature classification, and microstructural characterization using MR diffusional variance decomposition (DIVIDE) imaging. Eur Radiol 2021;31:8197-207. [Crossref] [PubMed]
31. Xie SH, Lang R, Li B, Zhao H, Wang P, He JL, Ma XY, Wu Q, Wang SY, Zhang HP, Gao Y, Wu JL. Evaluation of diffuse glioma grade and proliferation activity by different diffusion-weighted-imaging models including diffusion kurtosis imaging (DKI) and mean apparent propagator (MAP) MRI. Neuroradiology 2023;65:55-64. [Crossref] [PubMed]
32. Wang P, Huang H, Yang L, Chen G. The value of morphological magnetic resonance imaging features combined with diffusion kurtosis imaging for predicting Ki-67 expression levels in high-grade gliomas. Quant Imaging Med Surg 2025;15:2813-26. [Crossref] [PubMed]
33. Wang X, Li F, Wang D, Zeng Q. Diffusion kurtosis imaging combined with molecular markers as a comprehensive approach to predict overall survival in patients with gliomas. Eur J Radiol 2020;128:108985. [Crossref] [PubMed]
34. Genç B, Aslan K, Özçağlayan A, İncesu L. The role of MR diffusion kurtosis and neurite orientation dispersion and density imaging in evaluating gliomas. J Neuroimaging 2023;33:644-51. [Crossref] [PubMed]
35. Ke X, Zhao J, Liu X, Zhou Q, Cheng W, Zhang P, Zhou J. Apparent diffusion coefficient values effectively predict cell proliferation and determine oligodendroglioma grade. Neurosurg Rev 2023;46:83. [Crossref] [PubMed]
36. Wang S, Kim S, Chawla S, Wolf RL, Knipp DE, Vossough A, O'Rourke DM, Judy KD, Poptani H, Melhem ER. Differentiation between glioblastomas, solitary brain metastases, and primary cerebral lymphomas using diffusion tensor and dynamic susceptibility contrast-enhanced MR imaging. AJNR Am J Neuroradiol 2011;32:507-14. [Crossref] [PubMed]
37. Wang S, Kim S, Chawla S, Wolf RL, Zhang WG, O'Rourke DM, Judy KD, Melhem ER, Poptani H. Differentiation between glioblastomas and solitary brain metastases using diffusion tensor imaging. Neuroimage 2009;44:653-60. [Crossref] [PubMed]
38. Dehmeshki J, Barker GJ, Tofts PS. Classification of disease subgroup and correlation with disease severity using magnetic resonance imaging whole-brain histograms: application to magnetization transfer ratios and multiple sclerosis. IEEE Trans Med Imaging 2002;21:320-31. [Crossref] [PubMed]
39. Tozer DJ, Jäger HR, Danchaivijitr N, Benton CE, Tofts PS, Rees JH, Waldman AD. Apparent diffusion coefficient histograms may predict low-grade glioma subtype. NMR Biomed 2007;20:49-57. [Crossref] [PubMed]
40. Alexiou GA, Zikou A, Tsiouris S, Goussia A, Kosta P, Papadopoulos A, Voulgaris S, Kyritsis AP, Fotopoulos AD, Argyropoulou MI. Correlation of diffusion tensor, dynamic susceptibility contrast MRI and (99m)Tc-Tetrofosmin brain SPECT with tumour grade and Ki-67 immunohistochemistry in glioma. Clin Neurol Neurosurg 2014;116:41-5. [Crossref] [PubMed]
41. Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, Ohgaki H, Wiestler OD, Kleihues P, Ellison DW. The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol 2016;131:803-20. [Crossref] [PubMed]
42. Nielsen LAG, Bangsø JA, Lindahl KH, Dahlrot RH, Hjelmborg JVB, Hansen S, Kristensen BW. Evaluation of the proliferation marker Ki-67 in gliomas: Interobserver variability and digital quantification. Diagn Pathol 2018;13:38. [Crossref] [PubMed]
43. Xue B, Hou Z, Deng Z, Sun S, Zhang C, Pan Y, Zhang Y, Li Z, Xie J. Survival outcome and predictors of WHO grade 2 and 3 insular gliomas: A classification based on the tumor spread. Cancer Med 2024;13:e7377. [Crossref] [PubMed]
44. Yuan J, Xie D, Fang S, Meng F, Wu Y, Shan D, Shao N, Wang B, Tian Z, Wang Y, Xu C, Chen X. Qualitative and quantitative MRI analysis of alveolar soft part sarcoma: correlation with histological grade and Ki-67 expression. Insights Imaging 2024;15:142. [Crossref] [PubMed]
45. Hua B, Yang G, An Y, Lou K, Chen J, Quan G, Yuan T. Clinical and Imaging Characteristics of Contrast-enhanced Mammography and MRI to Distinguish Microinvasive Carcinoma from Ductal Carcinoma In situ. Acad Radiol 2024;31:4299-308. [Crossref] [PubMed]
46. Feng Z, Li H, Liu Q, Duan J, Zhou W, Yu X, Chen Q, Liu Z, Wang W, Rong P. CT Radiomics to Predict Macrotrabecular-Massive Subtype and Immune Status in Hepatocellular Carcinoma. Radiology 2023;307:e221291. [Crossref] [PubMed]
47. Kim SY, Cho N, Choi Y, Lee SH, Ha SM, Kim ES, Chang JM, Moon WK. Factors Affecting Pathologic Complete Response Following Neoadjuvant Chemotherapy in Breast Cancer: Development and Validation of a Predictive Nomogram. Radiology 2021;299:290-300. [Crossref] [PubMed]
48. Chu JP, Song YK, Tian YS, Qiu HS, Huang XH, Wang YL, Huang YQ, Zhao J. Diffusion kurtosis imaging in evaluating gliomas: different region of interest selection methods on time efficiency, measurement repeatability, and diagnostic ability. Eur Radiol 2021;31:729-39. [Crossref] [PubMed]
49. Zhao J, Wang YL, Li XB, Hu MS, Li ZH, Song YK, Wang JY, Tian YS, Liu DW, Yan X, Jiang L, Yang ZY, Chu JP. Comparative analysis of the diffusion kurtosis imaging and diffusion tensor imaging in grading gliomas, predicting tumour cell proliferation and IDH-1 gene mutation status. J Neurooncol 2019;141:195-203. [Crossref] [PubMed]
50. Andica C, Kamagata K, Hatano T, Saito Y, Ogaki K, Hattori N, Aoki S. MR Biomarkers of Degenerative Brain Disorders Derived From Diffusion Imaging. J Magn Reson Imaging 2020;52:1620-36. [Crossref] [PubMed]
51. Patel KS, Yao J, Cho NS, Sanvito F, Tessema K, Alvarado A, Dudley L, Rodriguez F, Everson R, Cloughesy TF, Salamon N, Liau LM, Kornblum HI, Ellingson BM. pH-Weighted amine chemical exchange saturation transfer echo planar imaging visualizes infiltrating glioblastoma cells. Neuro Oncol 2024;26:115-26. [Crossref] [PubMed]
52. Papacocea SI, Vrinceanu D, Dumitru M, Manole F, Serboiu C, Papacocea MT. Molecular Profile as an Outcome Predictor in Glioblastoma along with MRI Features and Surgical Resection: A Scoping Review. Int J Mol Sci 2024;25:9714. [Crossref] [PubMed]
53. Alnahhas I. Molecular Testing in Gliomas: What is Necessary in Routine Clinical Practice? Curr Oncol Rep 2024;26:1277-82. [Crossref] [PubMed]
54. Segura-Collar B, Hiller-Vallina S, de Dios O, Caamaño-Moreno M, Mondejar-Ruescas L, Sepulveda-Sanchez JM, Gargini R. Advanced immunotherapies for glioblastoma: tumor neoantigen vaccines in combination with immunomodulators. Acta Neuropathol Commun 2023;11:79. [Crossref] [PubMed]
55. Kim H, Yoon H, Thakur N, Hwang G, Lee EJ, Kim C, Chong Y. Deep learning-based histopathological segmentation for whole slide images of colorectal cancer in a compressed domain. Sci Rep 2021;11:22520. [Crossref] [PubMed]