The value of nomogram based on MRI functional imaging in differentiating cerebral alveolar echinococcosis from brain metastases

Abstract

Objective

This study aims to evaluate the effectiveness of a nomogram model constructed using Diffusion Kurtosis Imaging (DKI) and 3D Arterial Spin Labeling (3D-ASL) functional imaging techniques in distinguishing between cerebral alveolar echinococcosis (CAE) and brain metastases (BM).

Methods

Prospectively collected were 24 cases (86 lesions) of patients diagnosed with CAE and 16 cases (69 lesions) of patients diagnosed with BM at the affiliated hospital of Qinghai University from 2018 to 2023, confirmed either pathologically or through comprehensive diagnosis. Both patient groups underwent DKI and 3D-ASL scanning. DKI parameters (Kmean, Dmean, FA, ADC) and cerebral blood flow (CBF) were analyzed for the parenchymal area, edema area, and symmetrical normal brain tissue area in both groups. There were 155 lesions in total in the two groups of patients. We used SPSS to randomly select 70% as the training set (108 lesions) and the remaining 30% as the test set (47 lesions) and performed a difference analysis between the two groups. The independent factors distinguishing CAE from BM were identified using univariate and multivariate logistic regression analyses. Based on these factors, a diagnostic model was constructed and expressed as a nomogram.

Result

Univariate and multivariate logistic regression analyses identified nDmean1 and nCBF1 in the lesion parenchyma area, as well as nKmean2 and nDmean2 in the edema area, as independent factors for distinguishing CAE from BM. The model's performance, measured by the area under the ROC curve (AUC), had values of 0.942 and 0.989 for the training and test sets, respectively. Calibration curves demonstrated that the predicted probabilities were highly consistent with the actual values, and DCA confirmed the model's high clinical utility.

Conclusion

The nomogram model, which incorporates DKI and 3D-ASL functional imaging, effectively distinguishes CAE from BM. It offers an intuitive, accurate, and non-invasive method for differentiation, thus providing valuable guidance for subsequent clinical decisions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40001-024-02064-3.

Introduction

Echinococcosis is a zoonotic disease caused by tapeworms from the genus Echinococcus, which belongs to the family Cestodae [1]. Echinococcosis is divided into cystic echinococcosis (CE) and alveolar echinococcosis (AE), caused by the tapeworms Echinococcus granulosus and Echinococcus multilocularis, respectively [2]. Large-scale screening with ultrasound (US) and serology in China has confirmed a high incidence of AE in the Qinghai–Tibet Plateau, including Qinghai, Sichuan, and Tibet. An epidemiological survey of the Tibet Autonomous Region (TAR) revealed that the prevalence of human echinococcosis was 1.66%, which was much higher than the average prevalence in China (0.24%) [3]. The primary site of AE is usually the liver, where it grows in an exogenous, tumor-like manner with destructive proliferation. The larvae may spread locally or metastasize to other organs via blood or lymphatic pathways, resembling the behavior and appearance of malignant tumors [4]. Brain involvement in AE is rare, occurring in about 1% of cases, and typically manifests as focal neurological deficits and increased intracranial pressure [5]. The mortality and recurrence rates of cerebral alveolar echinococcosis (CAE) are high, and the treatment outcomes for CAE are also not satisfactory [6]. Brain metastases (BM) constitute about 50% of supratentorial brain tumors and are the most common type of secondary malignant brain tumor. BM frequently occurs in patients with lung cancer, breast cancer, and melanoma [7, 8].CAE and BM have similar symptoms and imaging manifestations. In daily clinical practice, when clinical information is limited, it is difficult for radiologists to accurately distinguish them based on conventional imaging features alone [6, 9].

Magnetic resonance imaging (MRI) is currently considered one of the most effective diagnostic tools for assessing the nervous system. CAE and BM demonstrate similar imaging characteristics, complicating their differentiation using standard imaging modalities like MRI and computed tomography (CT). Both conditions often present as solid-enhancing lesions with irregular or nodular patterns and can appear in any part of the brain. These lesions may show rim or heterogeneous enhancement, signaling active disease processes. Additionally, perilesional edema is frequently observed, a result of blood–brain barrier disruption and an accompanying inflammatory response. Consequently, accurate diagnosis through these routine imaging modalities is challenging. Diffusion kurtosis imaging (DKI) represents a new model based on the non-Gaussian distribution of water molecule diffusion. It detects subtle changes at the molecular level within brain tissue and provides multiple quantitative parameters that reflect the microscopic pathological changes in diseased tissue in vivo, such as inflammatory cell proliferation, cell edema or necrosis [10]. 3D-ASL (3D arterial spin labeling, ASL) is a non-invasive perfusion-weighted imaging technique that measures cerebral blood flow (CBF) by labeling the spins of flowing arterial blood, without the need for exogenous contrast agents [11, 12]. Currently, there is no report on the application of DKI and 3D-ASL technologies in the diagnosis and differential diagnosis of CAE. In previous studies, some scholars have used radiomics combined with machine learning to distinguish CAE from BM [13]. This study aims to evaluate the effectiveness of a multi-parameter nomogram model constructed using DKI and 3D-ASL functional imaging technologies in distinguishing CAE from BM.

We present the following article by the TRIPOD Guidelines reporting checklist.

Materials and methods

General information

This retrospective study, conducted by the Declaration of Helsinki, was approved by the local ethics committee, and informed consent was waived (P-SL-2023-229). From 2018 to 2023, patients diagnosed with CAE or BM at the Affiliated Hospital of Qinghai University were prospectively collected. The study included 24 patients with CAE (86 lesions) and 16 patients with BM (69 lesions). The inclusion criteria were: (1) CAE or BM confirmed by pathology; (2) CAE diagnosed by pathologically confirmed alveolar echinococcosis along with a comprehensive clinical diagnosis; (3) availability of multi-parametric MRI scan data from DKI and 3D-ASL; (4) no history of preoperative treatment; (5) all BM patients had a clear history of extracerebral malignant tumors; and (6) none of the BM patients had a history of brain tumors. The exclusion criteria included: (1) patients who had received any treatment for CAE or BM, such as surgery, radiotherapy, or chemotherapy; and (2) patients with unavailable or unqualified imaging data. All patients in this study had a total of 155 lesions. We used SPSS to randomly select 70% as the training set and the remaining 30% as the test set. (training set: 108 lesions, testing set: 47 lesions; within the training set, 59 lesions were CAE and 49 were BM; within the testing set, 27 lesions were CAE and 20 were BM).

Image acquisition

The imaging for all subjects in this study was performed using a Siemens Prisma 3.0 T MRI scanner with a head coil. The scanning sequences included: (1) Axial T2WI: TR 5000 ms, TE 117 ms, slice thickness 5.0 mm, FOV 230 mm, matrix 224 × 320. (2) Fluid-attenuated inversion recovery (FLAIR) sequence: TR 8000 ms, TE 81 ms, slice thickness 5.0 mm, FOV 230 mm, matrix 224 × 320. (3) DWI sequence: TR 3230 ms, TE 51 ms, slice thickness 5.0 mm, FOV 230 mm, matrix 160 × 160. (4) DKI sequence: TR 3000 ms, TE 68 ms, slice thickness 2.2 mm, FOV 220 mm, matrix 63 × 64. (5) 3D-ASL sequence: TR 3500 ms, TE 15.54 ms, slice thickness 3 mm, FOV 202 mm, matrix 63 × 64. (6) Axial enhanced T1WI: TR 2000 ms, TE 2.44 ms, slice thickness 1.0 mm, FOV 230 mm, matrix 224 × 224. The scans started with a plain scan followed by an enhanced scan, with 3D-ASL and DKI images acquired before the enhanced scan. The contrast agent used was Gadoteric Acid Meglumine Salt Injection, administered intravenously at a dose of 0.1 ml/kg body weight over 12 s, followed by a 20 ml saline flush.

Image post-processing

The number and location of CAE and BM lesions were observed and recorded on conventional MRI images. ADC Value Measurement: DWI images were imported into the workstation to generate ADC maps. Regions of Interest (ROIs) were delineated in the CAE and BM lesion parenchyma, the edema areas, and the contralateral normal tissue to measure ADC values. DKI Post-processing: The original DKI images of all patients were exported in DICOM format. The Diffusional Kurtosis Estimator (DKE, NITRC) software was utilized to calculate and create pseudo-color images for various parameters, including the mean kurtosis (Kmean) map, mean diffusion coefficient (Dmean) map, and fractional anisotropy (FA) map (see Figs. 1 and 2). 3D-ASL post-processing: The original 3D-ASL images of all patients were processed using Matlab 2016 software to generate cerebral blood flow (CBF) maps. Using conventional MRI and MRIcro software, ROIs were outlined in the parenchymal, edema, and contralateral normal tissue areas of CAE and BM lesions on the Kmean map, each covering an area of (10 ± 5) mm². These ROIs were then horizontally replicated to other parameter pseudo-color maps to derive values for each parameter. All quantitative data were measured and read three times by two radiologists, each with over 5 years of experience in MRI imaging diagnostics, and the average values were taken. Finally, the quantitative parameters of the parenchymal and edema areas of the lesions were compared with those of the contralateral normal tissue to derive standardized parameter values. The intraclass correlation coefficient (ICC) was employed to assess the consistency of the measurements between two physicians. An ICC greater than 0.75 indicated good consistency. The average measurement values from the two physicians were then used for further analysis.

Fig. 1.

Male, 34 years old, cerebral metastasis of hepatic alveolar echinococcosis. Panels a–d are the T1WI, T2WI, ADC, and DWI images of the CAE patient, respectively; panels e–h are the Kmean, Dmean, FA, and CBF images, respectively

Fig. 2.

Male, 43 years old, brain metastasis from lung cancer. Panels a–d are the T1WI, T2WI, ADC, and DWI images of the BM patient, respectively; panels e–h are the Kmean, Dmean, FA, and CBF images respectively

Construction and evaluation of diagnostic prediction models

In this study, a nomogram model was constructed using a training set. Single-factor and multi-factor logistic regression analyses were applied to the parameter characteristics of DKI and 3D-ASL to identify statistically significant indicators for distinguishing between CAE and BM lesions. These analyses estimated the odds ratio (OR) of each candidate variable, which was then used to build the model. A nomogram was subsequently created to visually represent the model.

The model's predictive ability was evaluated using the receiver-operating characteristic (ROC) curve, and the area under the curve (AUC) along with sensitivity and specificity were calculated based on the maximum Youden index. Calibration curves (CRC) were used to assess the model's goodness of fit, and the model was validated in the test set. Decision curve analysis (DCA) was used to quantify the net benefit to patients at different threshold probabilities, thereby assessing the clinical value of the model.

Statistical analysis

Statistical analyses were conducted using RStudio 4.2.3, Medcalc 20.022, and SPSS 25. In SPSS 25, the Shapiro–Wilk test was first used to assess the normality of continuous data distributions. Data conforming to normal distribution were expressed as $\overline{X}\pm \text{S}$, and comparisons between two groups were made using the independent sample t test; non-normally distributed data were analyzed using the rank-sum test, and represented by the interquartile range M (P75, P25). Categorical variables were represented by case numbers, and comparisons between two groups were made using the Chi-square test. Univariate analysis was conducted using the independent sample t test or Chi-square test to screen out factors with P < 0.05, which were then included in a multivariate logistic regression analysis. ROC curves were plotted using Medcalc. Calibration curves and nomograms were created using the ‘rms’ package in R, and decision curves were drawn using the ‘rmda’ package. A P value of < 0.05 was considered statistically significant.

Results

Clinical characteristics of patients

In this study, we collected data from 24 patients with CAE (86 lesions) and 16 patients with BM (69 lesions). The primary lesions in the 16 cases of BM included 9 cases of lung cancer, 1 case of breast cancer, 2 cases of liver cancer, 1 case of gastric cancer, 1 case of renal cancer, and 2 cases of malignant tumors in other locations. The clinical characteristics considered for both the CAE group and the BM group were age and gender. As shown in Table 1, there was no significant difference in gender between the CAE and BM cohorts; however, a statistically significant age difference was found (P < 0.05), with CAE being more common in young adults.

Table 1.

Comparison of general information between CAE and BM

	CAE	BM	Statistical value	P-value
Age	38.1 ± 13.1	58.9 ± 13.9	− 4.789	< 0.001
Sex			− 0.539	0.539
Male	17	7
Female	7	6

Comparison between training set and test set

In total, there were 155 lesions across the CAE and BM groups. The training set included 108 lesions, with 59 from CAE and 49 from BM; the test set included 47 lesions, with 27 from CAE and 20 from BM. According to Table 2, there was no statistically significant difference in the DKI and ASL parameters between the training and test sets (P > 0.05).

Table 2.

Comparison of DKI and ASL parameters of lesions in training and test sets

Parameters	Training sets	Test sets	Statistical value	P value
nKmean1	1.36 (1.14, 1.50)	1.32 (1.18, 1.45)	− 0.604	0.546
nDmean1	0.78 (0.73, 0.88)	0.76 (0.72, 0.85)	− 0.781	0.435
nFA1	0.46 (0.38, 0.55)	0.46 (0.37, 0.52)	− 0.707	0.480
nADC1	1.46 (1.28, 1.62)	1.48 (1.28, 1.63)	− 0.315	0.752
nCBF1	0.61 (0.50, 1.31)	0.63 (0.47, 1.46)	− 0.631	0.528
nKmean2	1.16 (0.90, 1.33)	1.12 (0.96, 1.23)	− 1.310	0.190
nDmean2	0.92 (0.81, 1.11)	0.91 (0.81, 1.06)	− 0.442	0.659
nFA2	0.69 ± 0.12	0.68 ± 0.11	− 0.328	0.743
nADC2	2.01 (1.90, 2.14)	1.99 (1.80, 2.20)	− 0.695	0.487
nCBF2	0.67 (0.59, 0.78)	0.67 (0.57, 0.81)	− 0.047	0.963

nKmean1: standardized mean kurtosis value after comparison of parenchymal area with symmetrical normal brain tissue. nKmean2: standardized mean kurtosis value after comparison of edema area with symmetrical normal brain tissue. nDmean1: standardized mean diffusion coefficient value after comparison of parenchymal area with symmetric normal brain tissue. nDmean2: standardized mean diffusion coefficient value after comparison of edema area with symmetric normal brain tissue. nFA1: standardized individual anisotropy score value after comparing parenchymal area with symmetric normal brain tissue. nFA2: standardized anisotropy score value for each of the edema regions in comparison with symmetrical normal brain tissue. nADC1: standardized ADC value after comparison of parenchymal area with symmetrical normal brain tissue. nADC2: standardized ADC value after comparison of edematous area with symmetrical normal brain tissue. nCBF1: standardized cerebral blood flow values after comparison of parenchymal regions with symmetrical normal brain tissue. nCBF2: standardized cerebral blood flow values after comparison of edematous areas with symmetrical normal brain tissue

Consistency assessment and collinearity diagnosis between feature groups

The ICC was used to evaluate the consistency of the parameters measured by the two physicians for DKI and 3D-ASL images. The ICC values ranged from 0.82 to 0.96. Since ICC values greater than 0.75 indicate good consistency, the average measurements taken by the two physicians were used for further analysis. Additionally, collinearity statistics were performed on all factors, and a Variance Inflation Factor (VIF) less than 5 indicated no collinearity among the factors.

Establishment of the nomogram diagnostic model

The model was constructed using data from the training set. Initial univariate analysis revealed that several standardized values were statistically significant between CAE and BM: mean kurtosis values of the parenchyma (nKmean1) and edema areas (nKmean2), mean diffusion coefficient values of the parenchyma (nDmean1) and edema areas (nDmean2), anisotropy score values of the parenchyma (nFA1) and edema areas (nFA2), the ADC value of the edema area (nADC2), and cerebral blood flow values of the parenchyma (nCBF1) and edema areas (nCBF2), all with P < 0.05 (Table 3). Subsequently, these factors were included in a multivariate logistic regression analysis, which identified four independent predictors—nDmean1 (OR 33264505.540, 95%CI 285.325–3.878E + 12, P = 0.004) and nCBF1 (OR 0.093, 95%CI 0.012–0.696, P = 0.021) from the parenchyma, and nKmean2 (OR 18605.500, 95%CI 6.012–57,583,940.34, P = 0.017) and nDmean2 (OR 0.005, 95%CI 0.000–0.779, P = 0.040) from the edema area—for distinguishing CAE from BM (Table 4, all P < 0.05). Based on these predictors, a diagnostic model was developed, and a nomogram was created using the R language, which included a green line representing the 95% confidence interval for each predictor. The total score, derived from the sum of points for each test variable, corresponded to the probability of CAE risk, as detailed in Fig. 3.

Table 3.

Univariate analysis of CAE and BM parameters in training set

Parameters	Training set		Statistical value	P value
	CAE	BM
nKmean1	1.45 (1.32, 1.52)	1.20 (1.05, 1.35)	− 5.086	< 0.001
nDmean1	0.76 (0.71, 0.80)	0.86 (0.77, 0.95)	− 4.546	< 0.001
nFA1	0.41 (0.37, 0.50)	0.51 (0.43, 0.58)	− 3.437	0.001
nADC1	1.46 (1.27, 1.58)	1.46 (1.32, 1.66)	− 0.963	0.336
nCBF1	0.51 (0.44, 0.57)	1.44 (1.02, 1.93)	− 8.373	< 0.001
nKmean2	1.30 (1.18, 1.37)	0.86 (0.77, 1.06)	− 7.309	< 0.001
nDmean2	0.84 (0.78, 0.90)	1.11 (1.02, 1.26)	− 6.45	< 0.001
nFA2	0.65 ± 0.09	0.73 ± 0.13	3.475	0.001
nADC2	1.98 ± 0.17	2.07 ± 0.27	2.153	0.034
nCBF2	0.72 (0.65, 0.80)	0.60 (0.57, 0.71)	− 3.838	< 0.001

nKmean1: standardized mean kurtosis value after comparison of parenchymal area with symmetrical normal brain tissue. nKmean2: standardized mean kurtosis value after comparison of edema area with symmetrical normal brain tissue. nDmean1: standardized mean diffusion coefficient value after comparison of parenchymal area with symmetric normal brain tissue. nDmean2: standardized mean diffusion coefficient value after comparison of edema area with symmetric normal brain tissue. nFA1: standardized individual anisotropy score value after comparing parenchymal area with symmetric normal brain tissue. nFA2: standardized anisotropy score value for each of the edema regions in comparison with symmetrical normal brain tissue. nADC1: standardized ADC value after comparison of parenchymal area with symmetrical normal brain tissue. nADC2: standardized ADC value after comparison of edematous area with symmetrical normal brain tissue. nCBF1: standardized cerebral blood flow values after comparison of parenchymal regions with symmetrical normal brain tissue. nCBF2: standardized cerebral blood flow values after comparison of edematous areas with symmetrical normal brain tissue

Table 4.

Multivariate analysis of CAE and BM parameters in training set

Parameters	OR	95%CI	P value
nKmean1	0.056	0.000–29.099	0.367
nDmean1	33,264,505.540	285.325–3.878E + 12	0.004
nFA1	0.005	0.000–3.508	0.113
nCBF1	0.093	0.012–0.696	0.021
nKmean2	18,605.500	6.012–57,583,940.34	0.017
nDmean2	0.005	0.000–0.779	0.040
nFA2	0.002	0.000–3.541	0.101
nADC2	14.560	0.185–1142.992	0.229
nCBF2	1.583	0.001–2049.927	0.900

nKmean1: standardized mean kurtosis value after comparison of parenchymal area with symmetrical normal brain tissue. nKmean2: standardized mean kurtosis value after comparison of edema area with symmetrical normal brain tissue. nDmean1: standardized mean diffusion coefficient value after comparison of parenchymal area with symmetric normal brain tissue. nDmean2: standardized mean diffusion coefficient value after comparison of edema area with symmetric normal brain tissue. nFA1: standardized individual anisotropy score value after comparing parenchymal area with symmetric normal brain tissue. nFA2: standardized anisotropy score value for each of the edema regions in comparison with symmetrical normal brain tissue. nADC1: standardized ADC value after comparison of parenchymal area with symmetrical normal brain tissue. nADC2: standardized ADC value after comparison of edematous area with symmetrical normal brain tissue. nCBF1: standardized cerebral blood flow values after comparison of parenchymal regions with symmetrical normal brain tissue. nCBF2: standardized cerebral blood flow values after comparison of edematous areas with symmetrical normal brain tissue

Fig. 3.

Nomogram of the differential diagnosis model between CAE and BM

Evaluation of the nomogram diagnostic model

In the training set, the largest area under the curve (AUC) for distinguishing CAE from BM was 0.942 (95% CI 0.896–0.988), achieved at a cut-off value of 0.8266, with a sensitivity of 87.21% and a specificity of 95.45%. In the test set, the highest AUC was 0.989 (95% CI 0.966–1.000), reached at a cut-off value of 0.9500, with a sensitivity of 100.00% and a specificity of 95.00%, as illustrated in Fig. 4. The calibration curves, used to evaluate the predictive performance of the model, showed mean absolute errors of 0.039 and 0.030 in the training and test sets, respectively, demonstrating good agreement between predicted values and actual clinical outcomes (Fig. 5). The model's clinical utility was further validated through decision curve analysis, which indicated significant clinical benefits when used within a certain range to differentiate CAE from BM, underscoring the high clinical applicability of this diagnostic model (refer to Fig. 6 for details).

Fig. 4.

The ROC curve of training and test set

Fig. 5.

The calibration curve of training and test set

Fig. 6.

The DCA of training and test set

Discussion

Existing literature on CAE is mainly case reports, lacking comprehensive research [5, 14–16]. Previous studies have indicated that CAE and BM often present with similar symptoms and imaging characteristics, yet their treatment strategies and prognoses differ significantly. CAE typically requires a combination of long-term antiparasitic medication and surgical resection for treatment. In contrast, BM is managed with a multidisciplinary approach that may include surgical resection, radiotherapy, systemic chemotherapy, or targeted therapy, depending on the type of primary cancer. Generally, BM is associated with advanced cancer stages and tends to have a poor prognosis. On the other hand, CAE progresses slowly, and with timely diagnosis and treatment, its prognosis can be substantially improved. Furthermore, performing a biopsy on CAE poses a risk of spreading parasites within the brain, making accurate diagnosis crucial to avoid unnecessary procedures [13]. However, CAE and BM present similar symptoms and imaging findings, including neurological issues like epilepsy, headaches, focal neurological deficits, and mental disorders. Both conditions may appear as multiple ring-enhancing masses with surrounding edema on imaging exams. This similarity often challenges radiologists in making an accurate diagnosis [6, 9]. Therefore, it is crucial to develop a method that can help radiologists accurately differentiate between CAE and BM. In previous studies, some scholars have used radiomics combined with machine learning to distinguish CAE from BM. They built five machine learning models. The AUC of support vector classifier (SVC), linear discrimination analysis (LDA), k-nearest neighbors (KNN), and Gaussian naïve bias (NB) algorithms in training (testing) sets are 0.99 (0.94), 1.00 (0.87), 0.98 (0.92), 0.97 (0.97), and 0.98 (0.93), respectively [13]. In this study, the multi-parameter nomogram model constructed using DKI and 3D-ASL functional imaging technology also obtained a higher AUC value. Through univariate and multivariate logistic regression analyses identified nDmean1 and nCBF1 in the lesion parenchyma area, as well as nKmean2 and nDmean2 in the edema area, as independent factors for distinguishing CAE from BM. The model's performance, measured by the area under the ROC curve (AUC), had values of 0.942 and 0.989 for the training and test sets, respectively. It can be seen that the AUC value in the test set is higher than in the previous studies, indicating that the model has a more stable performance. Calibration curves demonstrated that the predicted probabilities were highly consistent with the actual values, and DCA confirmed the model's high clinical utility.

DKI surpasses the conventional MRI sequences in reflecting tissue microstructure. While conventional MRI relies on ADC—the sole parameter of DWI—to provide quantitative information about diffusion rates, it offers only limited insights into tissue microstructure. In contrast, DKI is a refined technology that utilizes multiple b values and diffusion sensitization gradients in various directions. This approach delivers a range of quantitative parameters that accurately characterize the in vivo microenvironment from multiple perspectives. DKI enables more precise characterization of both water molecule diffusion and the complexity of tissue microstructures than conventional DWI [17–19]. The mean kurtosis (Kmean) represents the average value of diffusion kurtosis measured across all angles in three-dimensional space, indicating the complexity of tissue component structures. Higher values suggest more severe restrictions within the tissue structure and greater complexity in its components [20]. This study identifies nKmean2 in the lesion's edema area as an independent factor for distinguishing CAE from BM, whereas nKmean1 in the lesion's parenchyma area is not. We speculate that this difference may be attributed to the activity in the marginal zone of CAE, where vesicular proliferation and inflammatory cell proliferation lead to increased tissue complexity and consequently higher Kmean values. In contrast, the lesion's parenchyma area is largely inactive and necrotic, resulting in lower tissue complexity and, thus, lower Kmean values. The mean diffusion coefficient (Dmean) effectively reflects the restriction level of water molecule diffusion. By quantifying non-Gaussian water molecule diffusion, Dmean provides insights into microstructural and structural changes in tissues, showing high sensitivity to cellular edema and necrosis [21]. This study identified two key parameters—nDmean1 in the lesion's parenchyma area and nDmean2 in the edema area—as independent factors for distinguishing CAE from BM. The difference in microstructure between CAE and BM likely accounts for this finding. In CAE, the parenchyma area tends to have more calcium salt deposits, while the edema area may exhibit inflammatory cell proliferation and infiltration in the marginal zone, leading to more severe restrictions on the diffusion of water molecules. Additionally, the multivariate analysis indicated no statistically significant difference between the FA and ADC parameters for CAE and BM. This may be due to the insufficient sensitivity of the FA and ADC parameter characteristics in distinguishing between these two diseases.

CAE is a low-perfusion tumor. The parenchymal area is low-perfused. The reason is that there are multiple small vesicles inside CAE, which continue to grow and proliferate outward, requiring a lot of energy. Coagulative necrosis often occurs inside, so the internal blood supply is relatively insufficient, resulting in low perfusion [22]. BM is a high-perfusion tumor due to the action of endothelial growth factor, which induces tumor angiogenesis during its growth. This process destroys the blood–brain barrier and promotes vascularization, resulting in high-perfusion levels in the parenchymal area of the metastatic lesion [23]. ASL is a sensitive method for detecting tumor vascularity [24]. This study utilized 3D-ASL functional imaging technology and found that the normalized cerebral blood flow (nCBF1) in the lesion's parenchymal area is an independent factor for distinguishing CAE from BM. This finding aligns with the previous research, confirming that CAE and BM can be differentiated based on the CBF value in the parenchymal area.

The nomogram is a visualization tool that optimizes statistical models. It offers patients personalized and accurate risk assessments by integrating multiple predictive factors, enhancing the readability of research results. Its widespread use in clinical practice highlights its effectiveness [25–28]. This study utilized DKI and 3D-ASL imaging technologies to incorporate parameter characteristics into a model. Through multivariate logistic regression analysis, four main independent factors were identified that differentiate CAE from BM: nDmean1 and nCBF1 in the lesion parenchyma, and nKmean2 and nDmean2 in the lesion edema area. A differential diagnosis model for CAE and BM was subsequently constructed and presented as a nomogram. The nomogram's high predictive performance was confirmed by a Receiver-Operating Characteristic (ROC) curve, which showed an Area Under the Curve (AUC) of 0.942 in the training set. The model's accuracy was further validated through a calibration curve, demonstrating high consistency between predicted and measured values. This accuracy was confirmed in the test set. The clinical applicability of the nomogram was established through decision curve analysis, indicating its significant potential in non-invasively differentiating CAE from BM, thereby enabling the formulation of personalized treatment plans and maximizing patient benefits.

This is the first study to employ a multi-parameter nomogram model based on DKI and 3D-ASL functional imaging to differentiate CAE from BM. The nomogram we developed can identify CAE and BM non-invasively, avoiding the risk of lesions spreading to surrounding tissues or causing allergic reactions in patients due to pathological biopsy. Its purpose is to help clinicians make appropriate clinical treatment decisions based on individual circumstances.

Limitation

However, our study does have some limitations. First, due to the rarity of CAE, the sample size remains small despite nearly 5 years of data collection on CAE and BM. The small sample size may limit the ability to generalize the model to a larger population. We plan to conduct multicenter studies in the future to overcome this limitation. Multicenter studies can help address the issue of small sample size and enhance the generalizability of the research findings. Second, our model relies on the advanced imaging sequences of DKI and 3D-ASL, which may not be available in all institutions, potentially limiting the broader application of our model.

Conclusion

In conclusion, the multi-parameter nomogram model presented in this paper, which integrates DKI and 3D-ASL functional imaging technologies, exhibits impressive predictive accuracy and stability. This sophisticated model not only enhances the precision of preoperative differentiation between CAE and BM, but also provides a reliable, non-invasive diagnostic tool. By leveraging advanced imaging techniques and multi-parameter analysis, this model addresses a critical need for accurate pre-surgical assessment, offering substantial benefits to patients who require precise diagnoses prior to undergoing surgery. The ability to differentiate between CAE and BM with high accuracy helps ensure that patients receive the most appropriate and targeted treatment, thereby improving clinical outcomes and potentially reducing the need for invasive diagnostic procedures.

Author contributions

Guarantors of integrity of entire study, PQ.T, CY.L, HH.B; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, PQ.T, CY.L, SH.L, HH.B, MM.M, YJ.X, YA.DZ, XQ.Z; clinical studies, PQ.T, CY.L, HH.B; experimental studies, PQ.T, CY.L, HH.B; statistical analysis, PQ.T, CY.L, HH.B; and manuscript editing, PQ.T, CY.L, HH.B; PQ.T and CY.L contributed equally to this article.

Funding

This work was funded by National Key Clinical Specialty Construction Project of National Clinical Key Specialty [Office of Health Commission of Qinghai Province (2024) No. 90].

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This retrospective study, conducted in accordance with the Declaration of Helsinki, was approved by the local ethics committee, and informed consent was waived (P-SL-2023-229).

Consent for publication.

Not applicable.

Competing interests

The authors declare no competing interests.