Whole-tumor histogram analysis of synthetic MRI for discriminating benign from malignant sinonasal tumors: correlations with histopathologic factors

Abstract

Background

Histogram parameters from synthetic magnetic resonance imaging (SyMRI) may provide more diagnostic information than mean values in differentiating benign from malignant sinonasal tumors. The histopathologic basis of SyMRI in characterizing malignant sinonasal tumors is still unclear. This study aimed to explore the potential value of SyMRI quantitative maps with whole-lesion histogram analysis in the diagnosis of benign and malignant sinonasal tumors and the correlations between SyMRI-derived histogram metrics and histopathologic features in malignant sinonasal tumors.

Methods

A total of 76 patients (29 benign and 47 malignant) with sinonasal tumors were enrolled. Nine histogram parameters of the whole tumor were extracted from T1, T2, and proton density (PD) quantitative maps, respectively. Univariate and multivariate analyses were utilized to explore the association between benign and malignant sinonasal tumors. Models based on single, combined quantitative maps, and clinical features were established to evaluate the diagnostic performance. The Spearman correlation coefficient was used to assess the correlation between histogram quantitative metrics of SyMRI and histopathological features.

Results

For SyMRI parameters, 18 histogram metrics showed significant differences between benign and malignant sinonasal tumors (all P<0.05). The combined model based on T2 map (T2-90th percentile, Minimum, and Kurtosis) and clinical features (age and bone destruction) attained the best diagnostic performance in discrimination of benign and malignant sinonasal tumors with the highest area under the curve (AUC) of 0.908, sensitivity of 91.5%, and specificity of 82.8%. Moreover, several histogram quantitative parameters of malignancies were correlated with Ki-67 (r=−0.465 to −0.28), p53 (r=−0.476 to 0.414) and epidermal growth factor receptor (EGFR) status (r=−0.428/0.419). The T2-90th Percentile was independently associated with Ki-67 labeling index (LI) (P<0.05).

Conclusions

Whole-tumor histogram quantitative parameters of SyMRI could further improve the diagnostic performance in differentiating benign from malignant sinonasal tumors and may serve as potential biomarkers in assessing the histopathologic features.

Introduction

A wide spectrum of benign and malignant tumors can be encountered commonly in the sinonasal region and should be differentially identified in clinical practice (1,2). The clinical manifestations are often nonspecific and potentially overlapping between benign and malignant tumors (3,4). Hence, it is of paramount importance to differentiate benign tumor from malignancies for further therapeutic decisions and prognostic assessments. Currently, nasal endoscopy serves as the primary diagnostic modality for discriminating these two entities. Nonetheless, endoscopic examination only shows superficial tumor changes and cannot display the interior structure, which may be inconsistent with postoperative pathology (4,5). Computed tomography (CT) and conventional magnetic resonance imaging (cMRI) can not only reveal lesion boundaries and surrounding tissue invasion, but also offer some dependable information for treatment planning. Meanwhile, the morphologic characteristics provided by CT and cMRI are not always sufficient for evaluation of the biological properties of sinonasal tumors (6).

Synthetic magnetic resonance imaging (SyMRI) is an emerging imaging method, which employs a fast spin-echo multi-dynamic multi-echo (MDME) pulse sequence and generates multiple qualitative contrast (T1-, T2-weighted, etc.), as well as quantitative T1, T2, and proton density (PD) maps in a single scan (7,8). Compared to cMRI acquisition, SyMRI may be advantageous in that it provides qualitative and quantitative information simultaneously and has a shorter scan time (9). It has been widely applied in differential diagnosis and prognosis assessment in the field of head and neck, prostate, metastatic lymph nodes, and others (8,10-12). Notably, the mean value was the most widely used in the aforementioned studies; additional potential information hidden in the images may have been ignored (13,14).

Whole-tumor histogram analysis can provide more accurate and comprehensive information compared to a representative section. Meanwhile, histogram parameters reflect the spatial distribution of voxel gray-level intensity and provide accurate assessment of intra-tumor heterogeneity, which are invisible to radiologists (15,16). Previous studies have demonstrated that histogram parameters derived from SyMRI were associated with tumor progression, treatment responses, survival, and lymph node metastasis in breast cancer, prostate cancer, rectal cancer, nasopharyngeal carcinoma, and head and neck squamous cell carcinoma (HNSCC) (7,8,17-20). To date, and to the best of our knowledge, whole-tumor histogram parameters combined SyMRI have not been applied in sinonasal tumors. Furthermore, histopathological factors, including Ki-67, P53, and epidermal growth factor receptor (EGFR) status are linked to tumor proliferative activity, aggressiveness, and patient prognosis and have been demonstrated to be related to SyMRI quantitative metrics in the field of HNSCC (17) and nasopharyngeal carcinoma (8). Given that different histopathologic types of malignant sinonasal tumors show quite diverse prognoses, the preoperative prediction of histopathological status is crucial for treatment decision-making and prognostic assessment. Therefore, the purpose of this research was to explore the value of histogram analysis in differentiating the diagnostic performance of sinonasal tumors and to examine the correlation between histogram quantitative parameters derived from SyMRI and Ki-67, P53, and EGFR expression statuses. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2564/rc).

Methods

Study participants

This retrospective study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. It was approved by the Institutional Review Board of The Second Affiliated Hospital of Xi’an Jiaotong University (No. 2024-048). The requirement of informed consent was waived due to the retrospective nature of the study. In total, 92 patients with sinonasal tumors were recruited between March 2021 and March 2024 based on the following inclusion criteria: (I) no previous treatment; (II) histological confirmed sinonasal tumors (biopsy or surgery) in accordance with the guidelines outlined in the 5th Edition of the World Health Organization Classification of Head and Neck Tumors (2); (III) available complete sinonasal scan including SyMRI; and (IV) acceptable image quality for measurement. The exclusion criteria included the following: (I) history of prior chemotherapy and radiotherapy before magnetic resonance imaging (MRI) scan (n=2); (II) the tumor was too small (short-axis diameters less than 10 mm, n=5); (III) no pathological finding (n=6); and (IV) poor image quality due to severe artifacts (n=3). Ultimately, 76 patients with sinonasal tumors were included in the study. The workflow diagram of the study cohort is displayed in Figure 1.

Figure 1.

The workflow diagram of study population selection. MRI, magnetic resonance imaging; SyMRI, synthetic MRI.

Image acquisition

All sinonasal MRI examinations were performed on a 3.0 T MR scanner (SIGNA™ Architect, GE Healthcare, Milwaukee, WI, USA), with a 48-channel head coil. Detailed scan information is listed in Table S1. The following MR sequences were scanned: axial T1-weighted imaging, axial T2-weighted imaging, coronal T2-weighted imaging, and SyMRI (21).

Image processing and histogram analysis

The raw images were transformed to the dedicated Advantage Workstation (GE Healthcare, AW 4.7). Vendor provided post-processing programs (MAGiC, v. 100.1.1, GE Healthcare) were used to generate T1, T2, and PD maps from MAGiC sequence.

The cMRI images were used as a reference to delineate tumor boundaries clearly and accurately. Two radiologists (Reader 1 and Reader 2 with 5 and 10 years of experience in head and neck imaging diagnosis, respectively) independently manually outlined volumes of interest (VOIs) of whole tumors by using 3D slicer platform (version 5.2.2, https://www.slicer.org) (22), excluding any visible necrosis, cysts, and hemorrhage. Both readers were blinded to all the clinical information. Subsequently, the VOIs were copied and pasted onto the T1, T2, and PD maps. Nine first-order histogram parameters including the 10th percentile, 90th percentile, mean, median, maximum, minimum, skewness, kurtosis, and range values were extracted from SyMRI quantitative maps using Pyradiomics (version 2.0.0, https://www.radiomics.io/pyradiomics.html). In the end, 27 histogram parameters were obtained for each patient.

Two-dimensional region of interest segmentation of SyMRI

Reader 1, with 5 years of head and neck MRI experience, was engaged for the two-dimensional (2D) ROI segmentation analysis, who was blinded to the pathologic results. The 2D-ROIs were delineated to cover the largest area of tumor on the axial T1, T2, and PD maps using cMRI as a reference. Subsequently, mean quantitative parameters were obtained.

Histopathologic examination

The Ki-67, P53, and EGFR expression statuses of sinonasal malignant tumors were evaluated by a pathologist (with 6 years of experience in pathological tumor diagnosis), who was blinded to clinicopathologic results. The Ki-67 and P53 labeling index (LI) was (were is better? Two types of indicators) recorded based on the percentage of positively stained tumor cells and EGFR expression was assessed as the sum of stained areas (µm²). Ki-67 and P53 LI were divided into high expression (>50%) or low expression (≤50%), respectively (17). EGFR expression level was classified into high-expression (3+) and low expression (0–2+) (23).

Statistical analysis

All statistical analyses were performed using the software SPSS 26.0 (IBM Corp., Armonk, NY, USA) and GraphPad Prism (version 8.0, GraphPad Software, San Diego, CA, USA). Continuous data were presented as mean ± standard deviation or median (quartile), and categorical data were expressed as percentage. The interclass correlation coefficient (ICC) was used to evaluate the interobserver consistency (ICC ≥0.8, excellent; 0.6–0.8, good; 0.4–0.6, moderate; 0.2–0.4, poor) (24). Parameters were compared between benign and malignant sinonasal tumors using independent sample t-test or Mann-Whitney U test. A P value <0.05 was considered a statistical difference. Receiver operating characteristic (ROC) curves were performed for all significant parameters to investigate differential diagnostic efficiency. Then, Pearson/Spearman correlation coefficients (CCs) were calculated to evaluate the similarity of each significant histogram variable. The relevant parameters of each quantitative map were grouped according to which CCs were higher than 0.8, and the feature was retained with the highest performance [the highest area under the curve (AUC)] (20,25). Subsequently, the multivariate logistic regression analysis with a forward stepwise selection was performed to select and establish various diagnostic models. The Hosmer-Lemeshow test was used to assess the goodness-of-fit of the different diagnostic models. Meanwhile, the odds ratio (OR) and 95% confidence intervals (CIs) were recorded. To avoid overfitting, the selected characteristics from each single map were turned into a probability score. Thereafter, the probability scores were fed into a multivariate logistic regression model to identify the crucial probability score in a mixed feature set and construct a combined diagnostic model including clinical features. The diagnostic performance of single and combined models was assessed through ROC curves including the corresponding AUC, sensitivity, specificity, and Youden index. The Delong test was performed in comparison to the pairwise ROC curves.

Spearman rank correlation analysis was used to estimate the relationship between histopathologic factors and histogram quantitative parameters. A correlation coefficient rho (r) of 0.75–1.00 was considered a very good to excellent correlation; ≥0.7, strong correlation; 0.4–0.7, moderate correlation; and ≤0.4, mild correlation (8). The quantitative metrics with the maximum absolute r were included in stepwise multiple linear regression analyses to explore the associations between different models based on histogram parameters and histopathologic expression status, avoiding the multicollinearity between different imaging parameters.

Results

Population characteristics

A total of 76 patients were finally included in the study, and the details of clinical features are shown in Table 1. Among these patients, 29 were diagnosed with benign tumors, whereas 47 were diagnosed with malignant tumors. As displayed in Table 1, male patients accounted for a substantial part, although there was no statistical difference between the benign and malignant groups. Compared to patients with benign tumors, those with malignant tumors were older and had higher incidences of bone destruction (P<0.05).

Table 1. Clinical characteristics of patients.

Characteristic	Benign tumors (n=29)	Malignant tumors (n=47)	P value
Gender			0.183
Male	16 (55.17)	33 (70.21)
Female	13 (44.83)	14 (29.79)
Age (years)	48.69±12.72	60.26±16.98	0.002
Average diameter (cm)	3.91±1.11	4.22±1.18	0.257
Bone destruction	4 (13.79)	32 (68.09)	<0.001
Histologic subtypes, n	Inverted papilloma [11]	Squamous cell carcinoma [21]
	Hemangioma [9]	Lymphoma [11]
	Angiofibroma [2]	Malignant melanoma [6]
	Meningioma [2]	Adenoid cystic carcinoma [4]
	Ossifying fibroma [1]	Adenocarcinoma [3]
	Fibrous dysplasia of bone [1]	Neuroendocrine carcinomas [1]
	Pleomorphic adenoma [1]	Ewing’s sarcoma [1]
	Osteoma [1]
	Ameloblastoma [1]
Ki-67 expression
>50%		26 (55.32)
≤50%		21 (44.68)
P53 expression
>50%		20 (42.55)
≤50%		26 (55.32)
EGFR expression
3+		13 (27.66)
0–2+		13 (27.66)

Categorical variables are presented as n (%) and continuous variables are presented as mean ± standard deviation. Average diameters were measured at the maximum axial level of tumors. EGFR, epidermal growth factor receptor.

Interobserver agreement of quantitative parameters

In this study, all measurements of SyMRI histogram parameters showed excellent interobserver reproducibility and ranged from 0.801 to 0.985 (Table S2).

Models based on single functional map

The comparison of extracted histogram features from each quantitative map between benign and malignant tumors is displayed in Table S2 and Figure 2. Figure S1 shows the correlation of all significant parameters in function maps. For models derived from single quantitative map, 3, 3, and 3 histogram characteristics were chosen to construct diagnostic models, including T1 map (T1-Minimum, T1-Median, and T1-Kurtosis), T2 map (T2-90th percentile, T2-Minimum, and T2-Kurtosis), and PD map (PD-90th percentile, PD-Minimum, and PD-Median). As displayed in Figure 3, the values of T1-Minimum, T1-Median, T2-90th percentile, T2-Minimum, PD-90th percentile, PD-Minimum, and PD-Median were lower in malignant than benign sinonasal tumors. On the contrary, malignant tumors showed higher T1, T2-Kurtosis, and Skewness values, compared with benign ones. Multivariate logistic regression analyses of all significant parameters are presented in Table 2. The T2 map-derived model showed the best diagnostic performance, with an AUC of 0.842 among all models based on a single quantitative map (Table 3). The ROC analysis results of the three models are presented in Table 3 and Figure 4A.

Figure 2.

Representative SyMRI and whole-tumor histogram analyses. (A-E) Case 1. A 48-year-old female with sinonasal hemangioma. (F-J) Case 2. A 53-year-old female with nasal SCC, respectively. An ROI was placed along the lesion (red circle). Both patients demonstrated well-defined masses in the left nasal cavity. Hemangioma (A) showed hyperintensity and SCC (F) displayed heterogeneous isointensity on T2-weighted imaging of SyMRI. In comparison with hemangioma, SCC showed lower T1, T2, and PD value on T1 map (B,G), T2 map (C,H), and PD map (D,I), respectively. Manual three-dimensional segmentations of tumors (E,J), histograms of T1 map (K), T2 map (L) and PD map (M). PD, proton density; ROI, region of interest; SCC, squamous cell carcinoma; SyMRI, synthetic MRI.

Figure 3.

Column charts show the comparison of representative histogram features based on single SyMRI quantitative map in benign and malignant sinonasal tumors. T1 map (A), T2 map (B), and PD map (C). *, P<0.05; **, P<0.01; ***, P<0.001. PD, proton density; SyMRI, synthetic magnetic resonance imaging.

Table 2. Multivariate logistic regression parameters.

Models	OR	95% CI	P value	Model fit†
Whole-tumor histogram models based on single SyMRI quantitative map
T1 map				0.512
Minimum	1.010	1.004–1.017	0.003
Median	1.002	0.999–1.005	0.207
Kurtosis	0.740	0.568–0.964	0.026
T2 map				0.757
90th percentile	1.026	1.010–1.041	0.001
Minimum	1.064	1.010–1.122	0.020
Kurtosis	0.998	0.994–1.003	0.475
PD map				0.586
90th percentile	1.064	0.938–1.209	0.335
Minimum	1.046	1.005–1.09	0.029
Median	1.066	0.96–1.184	0.23
Models based on combined SyMRI quantitative maps and clinical feature
SyMRI map + clinical feature				0.357
T2 map (probability score)	201.311	11.636–3,482.868	<0.001
Age	1.011	0.965–1.061	0.638
Bone destruction	13.600	2.923–63.266	0.001

^†, the Hosmer-Lemeshow test was implemented to evaluate the goodness-of-fit of the multivariate logistic models. A model with P>0.05 was considered well fitted. CI, confidence interval; OR, odds ratio; PD, proton density; SyMRI, synthetic magnetic resonance imaging.

Table 3. Diagnostic performance of models based on single and combined functional parameters in differentiating benign from malignant sinonasal tumors.

Model	AUC (95% CI)	Accuracy (%)	Sensitivity (%)	Specificity (%)	P value
Models based on single SyMRI quantitative map
T1 map (T1-minimum, T1-median, and T1-kurtosis)	0.794 (0.682–0.906)	79.0	83.0	72.4	<0.001
T2 map (T2-90th percentile, T2-minimum, and T2-kurtosis)	0.842 (0.738–0.945)	82.9	78.7	89.7	<0.001
PD map (PD-90th percentile, PD-minimum, and PD-median)	0.769 (0.658–0.880)	75.0	78.7	69.0	<0.001
Combined model based on SyMRI quantitative map and clinical features
Combined model† (the probability scores of T2 map + age + bone destruction)	0.908 (0.835–0.981)	88.2	91.5	82.8	<0.001

^†, to avoid overfitting, the selected characteristics from single quantitative map were transformed into a probability score. The probability scores were then fed into the multivariate logistic regression model to choose the key probability score in a mixed feature set and establish a combined model with clinical features. AUC, area under the curve; CI, confidence interval; PD, proton density; SyMRI, synthetic magnetic resonance imaging.

Figure 4.

ROC curves of different models and correlation analyses. (A) Comparison of ROC curves of diagnostic models based on single and combined SyMRI quantitative maps. (B) Correlation between the T2-90th percentile and Ki-67 LI (r=−0.465, P<0.001). AUC, area under the curve; LI, labeling index; PD, proton density; ROC, receiver operating characteristic; SyMRI, synthetic magnetic resonance imaging.

Combined model based on SyMRI quantitative maps and clinical features

The probability scores of T1 map, T2 map, and PD map were included in multivariate logistic regression analyses, and the key probability score of T2 map (Table S3) was identified as independent risk factor and selected to construct diagnostic model combined with clinical features (age and bone destruction). In comparison with a single model, the combined model was confirmed to have the highest diagnostic efficiency (Figure 4A, AUC =0.908, Delong test: P<0.05).

The mean parameters derived from 2D-ROI segmentation of SyMRI in benign and malignant sinonasal tumors

As shown in Table S4, mean T1, T2, and PD values were significantly lower in the malignant group compared to the benign group, consistent with a previous study (12). The AUCs of T1, T2, and PD maps were 0.734, 0.773, and 0.668, respectively.

Correlation of histogram quantitative parameters with histopathologic features

The correlations of histogram metrics with histopathological status of sinonasal malignant tumors are shown in Table 4. The T1-10th percentile, 90th percentile, mean, median, T2-10th percentile, 90th percentile, mean, median, minimum, and PD-10th percentile, mean, and median values were all negatively correlated with Ki-67 LI. T1-Skewness was positively correlated with Ki-67 LI. The T1-10th percentile and minimum were positively correlated with P53 LI. Meanwhile, the T1-Range and PD-maximum and range were all negatively correlated with P53 LI. For EGFR expression, only T1-Minimum and range exhibited fair correlation. In the stepwise multiple linear regression analyses, only T2-90th percentile was independently associated with Ki-67 LI (standardized β coefficient =−0.465, P<0.001, Figure 4B). There was no independent correlation between histogram metrics and P53 and EGFR expression statuses (P>0.05).

Table 4. Correlations between histogram metrics derived from SyMRI and histopathological features.

Quantitative map	Parameters	Ki-67			P53			EGFR
		r	P value		r	P value		r	P value
T1 map	10th percentile	−0.304	0.019*		0.394	0.018*		0.166	0.31
	90th percentile	−0.276	0.034*		0.068	0.692		0.094	0.565
	Mean	−0.387	0.002*		0.14	0.416		0.103	0.528
	Median	−0.434	0.001*		0.2	0.243		0.085	0.602
	Maximum	0.075	0.574		−0.18	0.292		−0.239	0.202
	Minimum	−0.186	0.159		0.414	0.012*		0.419	0.012*
	Skewness	0.28	0.032*		−0.125	0.469		−0.13	0.427
	Kurtosis	0.081	0.541		−0.076	0.66		−0.183	0.261
	Range	0.107	0.421		−0.043	0.015*		−0.428	0.011*
T2 map	10th percentile	−0.412	0.001*		−0.168	0.264		−0.079	0.614
	90th percentile	−0.465	<0.001*		−0.137	0.365		0.03	0.848
	Mean	−0.442	<0.001*		−0.156	0.301		0.011	0.943
	Median	−0.46	<0.001*		−0.152	0.313		−0.049	0.755
	Maximum	0.016	0.907		−0.023	0.878		−0.164	0.3
	Minimum	−0.33	0.011*		−0.02	0.894		0.169	0.28
	Skewness	0.068	0.61		0.206	0.169		−0.131	0.401
	Kurtosis	−0.041	0.759		0.205	0.171		−0.184	0.24
	Range	0.022	0.868		−0.018	0.904		−0.191	0.221
PD map	10th percentile	−0.287	0.027*		0.079	0.647		0.063	0.701
	90th percentile	0.041	0.76		0.065	0.708		0.063	0.701
	Mean	−0.325	0.012*		0.059	0.733		0.076	0.641
	Median	−0.341	0.008*		0	0.998		0.108	0.51
	Maximum	0.07	0.6		−0.476	0.003*		−0.115	0.49
	Minimum	−0.139	0.293		0.182	0.288		0.201	0.217
	Skewness	0.225	0.087		0.224	0.19		0.166	0.31
	Kurtosis	−0.198	0.134		−0.27	0.111		−0.237	0.146
	Range	0.122	0.357		−0.386	0.02*		−0.162	0.323

*, significant difference. EGFR, epidermal growth factor receptor; PD, proton density; SyMRI, synthetic magnetic resonance imaging.

In univariate analysis, the Ki-67 high expression group showed lower 10th percentile of T2 and PD, T2-90th percentile, and minimum values, including mean, median of T1, and T2 and PD, than the Ki-67 low expression group (all P<0.05). The diagnostic performance of histogram parameters from T1, T2, and PD maps was 0.682–0.718, 0.653–0.737, and 0.659–0.666 of AUC, respectively (Table S5). Meanwhile, T2-median exhibited the highest AUC of 0.737 (Figure 5A).

Figure 5.

ROC curves of representative histogram metrics of SyMRI in each map (T1 map, T2 map and PD map) in differentiating (A) Ki-67, (B) P53 and (C) EGFR expression status of sinonasal malignant tumors. AUC, area under the curve; EGFR, epidermal growth factor receptor; PD, proton density; ROC, receiver operating characteristic; SyMRI, synthetic magnetic resonance imaging.

The T1-90th percentile and PD-Kurtosis values were lower in the P53 high expression group than in the low expression group (all P<0.05). Furthermore, PD-Kurtosis displayed the best diagnostic performance (AUC =0.694, Figure 5B). Only T1-range was significantly different between the EGFR high- and low expression groups, with an AUC of 0.731 (Figure 5C).

Discussion

In this study, our data indicated that the SyMRI-derived whole-lesion histogram parameters models might be valuable to differentiate benign from malignant sinonasal tumors. Combined histogram characteristics based on SyMRI quantitative maps and clinical features showed optimal diagnostic capability with an AUC of 0.908. Moreover, the histogram quantitative metrics showed moderate-to-mild correlation in Ki-67, P53, and EGFR expression statuses. Compared with the reference standard of biopsy or surgery, the histogram parameters derived from SyMRI may serve as a noninvasive method to predict histopathologic factors for prognostic evaluation and clinical decision-making.

The T1-mean, T1-median, T1-minimum, T1-percentiles, T2-mean, T2-median, T2-minimum, T2-percentiles, PD-90th percentile, PD-mean, PD-median, and PD-minimum values of SyMRI quantitative parameters were significantly lower in malignant sinonasal tumors than they were in benign ones. The possible explanation might be that T1, T2, and PD relaxation times could reflect physical properties of tissue and link MRI images with tissue physiology (26). Moreover, T1, T2, and PD values mainly correlate with tissue composition, including macromolecule and water content (10). Malignant tumors are characterized by cellular necrosis, uncontrollable proliferation, cellular pleomorphism, angiogenesis, and so on (8,11). With the corresponding reduced extracellular fluid space, the T1, T2, and PD values might be decreased along with reduced free water content (10,27).

In a previous study, mean values of 2D-ROI quantitative parameters from SyMRI can serve as reliable noninvasive predictors for distinguishing between benign and malignant sinonasal lesions (12). Moreover, this research revealed higher AUCs of whole-tumor histogram models based on SyMRI parameters than results of 2D-ROI parameters (T1 map: 0.794 vs. 0.734, T2 map: 0.842 vs. 0.773, PD map: 0.769 vs. 0.668, Delong tests: P<0.05). These findings show that the mean values may not be the most appropriate metric for characterizing benign and malignant sinonasal tumors, which is attributed to the fact that mean values cannot not assess the deep complexity and heterogeneity of tumors (20,25). In our study, 27 first-order histogram parameters were extracted from SyMRI quantitative maps. Percentile parameters are less affected by random fluctuations than mean values (28) and may be especially adapted to evaluate the malignant components within lesions (29). In addition, skewness and kurtosis show the symmetry as well as peakedness of histogram, which could indirectly display the image gray-level heterogeneity (29). Therefore, whole-tumor histogram analysis yields many additional parameters which can provide more abundant information on tumor characteristics than conventional method (e.g., single parameter and single-slice) and better reflect different tumor microenvironments that may be masked by mean values (30).

Histogram analysis of SyMRI quantitative maps based on voxel distribution can eliminate sampling bias and represent intratumor heterogeneity more accurately, which has been confirmed in previous studies of HNSCC (17), nasopharyngeal carcinoma (23), rectal cancer (18), and breast cancer (7). Tumor heterogeneity was considered to result from regional variations in tumor cellularity, proliferation, angiogenesis, hypoxia, and necrosis, all of which were related to histopathological features (31). Presumably, Ki-67, P53, and EGFR expression statuses could reflect heterogeneity and aggressiveness of tumors quantitatively, which are always associated with the extent of proliferative activity and prognosis in various tumors. To date, the association of histogram parameters derived from SyMRI with histopathologic features in malignant sinonasal tumors has not been reported yet.

Ki-67 expression status reflects the proliferative activity of sinonasal tumors, as well as the high Ki-67 expression level is associated with a poor prognosis (32). Valente et al. (32) found that sinonasal carcinoma patients with Ki-67 LI (>50%) had a poorer prognosis than those with Ki-67 LI ≤50%. Our results demonstrated that histogram parameters, including percentiles, mean, median, minimum, and skewness were negatively correlated with Ki-67 expression levels. The Ki-67 high expression group of sinonasal malignant tumors was generally of increased tumor cellularity compared with the low-expression group, which could reduce extracellular fluid space and thus decrease the T1, T2, and PD values. Moreover, T1-Skewness values were positively correlated with Ki-67 LI, which also reflects tumor heterogeneity to some extent. Therefore, we speculated that sinonasal malignant tumors in the high Ki-67-expression group were more heterogeneous compared to those in the low-expression group. Furthermore, in stepwise multiple linear regression analysis, T2-90th percentile showed independent association with Ki-67 expression level of malignant sinonasal tumors. Thereby, SyMRI-derived histogram metrics can reflect cell proliferation status, which is of great clinical relevance for treatment decisions and outcome.

Currently, high-expression statuses of P53 and EGFR in tumors are always associated with poor prognosis (33,34). P53 regulates cell cycle arrest, DNA repair, senescence, or apoptosis (35). Another factor, namely, EGFR, is a biomarker that influences gene expression, cell proliferation, apoptosis, and metastasis of cancer (35). Meyer et al. elucidated the correlation between histogram parameters derived from dynamic-contrast-enhanced MRI (DCE-MRI) and P53 and EGFR expression in HNSCC (36). Meanwhile, Dang et al. demonstrated that MRI texture analysis predicts P53 expression status with 80% accuracy in oropharyngeal squamous cell carcinoma (37). These results indicated that imaging sequences might reflect tumor histopathology to some extent. Thus, the feasibility of preoperative assessment of histopathologic factors based on imaging is very important. In our study, regarding histogram quantitative metrics, moderate-to-mild association with P53 and EGFR expression was presented, in similarity with previous studies (36,37).

This study has several limitations. Firstly, the sample size was small, especially that of the patients with benign sinonasal tumors, which may be influenced by selection bias. The diagnostic performance of models may be affected by the imbalanced proportions of benign and malignant. Secondly, biopsy samples may not adequately represent entire tumor regions due to intra-tumoral heterogeneity of histopathologic factors expression. In this regard, we should include patients who have been identified by postoperative pathology in future research. Thirdly, the correlation between MRI images and histopathologic features was not observed on a site-to-site basis in the present study. This fact might limit our results. Then, despite the use of integrated MRI software for image alignment and realignment function in 3D Slicer, potential misalignment between the VOIs derived from cMRI and SyMRI may still exist. Lastly, our study only used single histogram analysis, which is unable to comprehensively evaluate tumor heterogeneity. Therefore, multiparametric MRI models may be optimized by incorporating texture analysis and machine learning techniques. Currently, artificial intelligence (AI) has demonstrated promising results in the application of radiomic analysis of MRI to assess sinonasal tumors (38). In the future, we will integrate AI with SyMRI and whole slide imaging (WSI) (39) to evaluate the prognostic factors in sinonasal tumors.

Conclusions

Whole-tumor histogram parameters derived from SyMRI quantitative maps of sinonasal tumors may be recognized as a noninvasive and free-contrast method for differentiation of benign and malignant tumors. Furthermore, there are multiple associations between histogram metrics derived from SyMRI and clinically relevant histopathologic features. Therefore, histogram quantitative parameters can be used as a potential biomarker for discriminating histopathologic features.

Supplementary

The article’s supplementary files as

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 and approved by the Institutional Review Board of The Second Affiliated Hospital of Xi’an Jiaotong University (No. 2024-048). The requirement of informed consent was waived due to the retrospective nature of the study.

Data Sharing Statement

Available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2564/dss

qims-15-08-7392-dss.pdf ^{(138.7KB, pdf)}

DOI: 10.21037/qims-2024-2564