Imaging prediction of malignant and benign neoplasms in the salivary gland using two-phase multislice spiral CT texture analysis

doi:10.21203/rs.3.rs-3134594/v1

Download PDF

Research Article

Imaging prediction of malignant and benign neoplasms in the salivary gland using two-phase multislice spiral CT texture analysis

https://doi.org/10.21203/rs.3.rs-3134594/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Parotid lesions show overlapping morphological findings and types of time/intensity curves. This research aimed to evaluate the role of two-phase multislice spiral computed tomography (MSCT) texture analysis in differentiating between benign and malignant parotid lesions.

Methods

In this retrospective research, MSCT was carried out on 90 patients. Each lesion was segmented on axial CT images by manually, and 33 texture features and morphologic CT features were assessed. Logistic regression analysis was used to identify predictors of malignancy (P < 0.05 was considered statistically significant), followed by receiver operating characteristic (ROC) analysis to determine diagnostic efficacy.

Results

Univariate logistic regression analysis demonstrated that morphological CT features (shape, size, and invasion of adjacent tissues) and 17 CT texture parameters were significant differences between benign and malignant lesions (P < 0.05). Multivariate binary logistic regression demonstrated that shape, invasion of adjacent tissues, entropy and inverse difference moment were independent factors for malignant tumours. The diagnostic accuracy of multivariate binary logistic models based on morphological parameters, CT texture features, and a combination of both was 87.8%, 90%, and 93.3%, respectively.

Conclusion

Two-phase MSCT texture analysis was conducive to differentiating between malignant and benign neoplasms in the salivary gland, especially when combined with morphological CT features.

Multislice spiral computed tomography (MSCT)

texture analysis

computer applications-detection/diagnosis

salivary gland neoplasms

Salivary gland tumours (SGTs) account for about 3%-5% of head and neck tumours, with nearly 15.7% being malignant[1, 2]. Both malignant and benign SGTs morphologic features exhibit abundant diversity and complexity, making it difficult to distinguish between them[1, 3]. The treatment of choice for most SGTs is surgery. However, a correct preoperative diagnosis of salivary gland tumours is especially important for policy of lymph node dissection, timely treatment to malignant tumours and possible operative risk assessment[4]. Moreover, early diagnosis can improves the prognosis of salivary gland tumours[4], it is crucial for the surgical treatment of malignant salivary gland tumours of low grade with a five-year survival rate of up to 90%[6]. The five-year survival rate in high-grade malignant SGTs patients reaches 50% [5, 6].

Clinically, CT is available to aid in the diagnosis of SGTs. However, CT imaging show a unreliable ability of differential diagnosis between malignant and benign SGTs due to restrictions in the histomorphology of tumours and the observer’s expertise[7, 8]. Therefore, an additional investigation is required to improve the diagnostic accuracy. CT texture analysis, a kind of technology in radiomics, can quantitatively analyse characteristics from CT imaging. The principle of MSCT texture analysis is to evaluate tumour heterogeneity by analysing the relationship and distribution of voxel or pixel grey levels in the CT image. Therefore, this is considered to be a more objective technique than traditional methods. CT texture analysis has been widely available in a variety of tumours and nononcologic applications, including head and neck cancer[9], liver cancer[10], lung cancer[11], quantification of hepatic fibrosis[12] and pulmonary fibrosis[13], in predicting pathologic features and assessment of therapy and prognosis. In addition, contrast-enhanced CT is also used in CT texture analysis, due to various blood supply characteristics of the tumour[14–16]. Previous studies also reported that two-phase CT could differentiate neoplasms in salivary glands[17–19]. Therefore, this research was intended to evaluate the diagnostic ability of texture parameters by a two-phase CT scan for the identification of salivary gland tumours (malignant and benign). Additionally, this research aimed to assess the diagnostic performance of morphology characteristics and a combination of morphology characteristics and two-phase CT texture features for the differentiation of salivary gland tumours.

2.1. Study Population

This was a retrospective research that included consecutive patients from January 2001 to August 2018. This study was authorized by Ethics Committee of the First Affiliated Hospital of Guangzhou University of Chinese Medicine (Guangzhou University of Chinese Medicine, Guangzhou, China), and patient informed consent was waived for this type of review. The inclusion criteria for salivary gland tumours were as follows: (i) a long axis diameter of > 1 cm to reliably obtain better statistics in terms of more voxels available within the neoplasm for CT texture analysis calculations; (ii) no operative treatment; (iii) tissue sampling performed only by postoperative pathological analysis; and (iv) tumours distorted by artefacts were excluded, as such artefacts may also disturb texture calculations. All patients underwent enhanced CT examination before surgery. At the same time, tree patients with obvious motion artefacts on CT images were excluded. A total of 90 patients with pathologically confirmed salivary gland neoplasms were analysed retrospectively. There were 43 women and 47 men, aged from 15 years to 87 years with a mean of 50.4 ± 18.0 years, who were included in this research.

2.2. CT imaging protocol

All patients included in this study underwent dynamic neck contrast-enhanced CT performed on a multiple-row detector 64 slices CT scanner (Siemens Healthcare, Erlangen, Germany). The details of CT acquisition parameters were as follows: tube voltage: 120–140 kV; tube current: 150–250 mAs/slice; gantry rotation time: 0.9 s; reconstruction thickness: 3 mm. For contrast-enhanced CT scanning, the patients were administered a total of 60 to 100 mL (1.25 mL.kg^− 1 of weight) of nonionic iodinated contrast agent (350 mg/mL; iohexol, GE Healthcare, Oslo, Norway). The contrast material was injected by using a power injector at a rate of 2.0 to 2.5 mL/s. CT examination coverage was conducted from the skull base to the superior aperture of thorax. Early phase and delayed phase scans were acquired with scanning delays of 30 s and 120 s, respectively.

2.3. CT qualitative image analysis

The lesions were assessed by two radiologists (both of whom had over 5 years of experience with head and neck tumours) in a blinded manner. Only one salivary gland neoplasm was selected for analysis in each patient. Conventional assessment indices for diagnosis included morphology (shape, size, margin and internal enhancement pattern), calcification, necrosis, cysts, lymphadenectasis and invasion of adjacent tissues. The shape was categorized as round, lobulated, and irregular. The size of tumour was measured in the maximum diameter in either sagittal or transverse planes. The margin of tumour was categorized as clear and undefined, and the internal enhancement mode was divided into homogeneous, heterogeneous or rim. The minimum diameter of lymph node enlargement cross-section is 10 mm as the evaluation standard.

2.4. CT texture analysis

All 90 patients’ CT data were divided into arterial and venous phases and analysed by using professional quantitative analysis software (Omni Kinetics, GE Healthcare). To improve the accuracy of the texture calculations, the region of interest (ROI) in the lesion was manually drawn on both arterial and venous phases by the two radiologists. The ROI in the lesion included vessels and solid components, except for necrosis, cystic degeneration and calcification(Fig. 1). Then, texture parameters calculated by software were acquired. These included the histogram parameters, grey-level cooccurrence matrix parameters (GLCM, second-order parameter), and grey-level run length maxia parameters (GLRLM, third-order parameters). A total of 33 CT texture parameters in a phase were extracted (their complete description is reported in Supplementary Table, available in repository data).

2.5. Statistical Analysis

Continuous variables were presented as the mean ± standard deviation. Differences between benign and malignant lesions were analysed using the χ² test (or Fisher’s exact test) for categorical variables and the Mann‒Whitney U test for continuous variables, including texture parameters. Univariate binary logistic regression analysis was applied to determine whether a single parameter could distinguish malignancy in SGTs patients. The diagnostic performance of each parameter was assessed using ROC analysis. The area under the ROC curve (AUC), specificity, sensitivity, negative predictive value (NPV), positive predictive value (PPV) and accuracy of each parameter were calculated and optimal cut-off values were determined using Youden's index. Those variables with P value less than 0.05, as determined by the univariate binary logistic regression analysis or by χ² test, were chosen for multivariate binary logistic regression analysis to determine the parameters that can best predict the diagnosis of malignancy. Odds ratios (ORs) with 95% confidence intervals (CIs) were used as relative risk estimates for each independent predictor. Three models were constructed using the multiple logistic regression analysis. In model 1, morphological parameters were selected as the independent risk factors; In model 2, CT texture parameters were elected as the independent risk factors; and In model 3, CT texture parameters combined with morphological parameters were screened as the independent risk factors. ROC analysis was used to evaluate the diagnostic performance of the three models, and the AUC of the three models were compared using the Hanley and McNeill tests. The specificity, sensitivity, NPV, PPV, and accuracy were calculated for each model. A two-tailed P value of less than 0.05 was considered statistically significant. All statistical analysis were performed by using SPSS (version 22.0; SPSS, Chicago, IL, USA).

3.1 Study population

Malignant tumours (n = 23) and benign tumours (n = 67) were enrolled in this study, including mucoepidermoid carcinoma (n = 5), adenoid cystic carcinoma (n = 5), salivary duct carcinoma (n = 4), lymphoma (n = 3), lympho-epithelioma (n = 2), malignant pleomorphic adenoma (n = 1), squamous carcinoma (n = 1), undifferentiated carcinoma (n = 1), adenocarcinoma (n = 1), pleomorphic adenoma (n = 35), Warthin tumour (n = 25) and base cell adenoma (n = 7).

3.2 Characteristics of lesions and statistical analysis

The characteristics of lesions and the comparison of benign and malignant lesions are summarized in Table 1. The mean size of malignant tumours was 43.26 ± 21.16 mm, which tended to be larger than that of benign tumours (26.06 ± 9.65 mm, P < 0.001). The shape of a benign tumour was more likely to be round or lobulated; however, malignant tumours were more irregular (P < 0.001). Moreover, the possibility of the appearance of an undefined border, calcification and invasion of adjacent tissues in malignant tissues was higher than that in benign tissues (P < 0.001). There was no statistic difference in necrosis, cysts, or enhancement pattern. A single parameter was capable to distinguish between malignant and benign lesions. As shown in Table 1, the mean size, shape, calcification and invasion of adjacent tissues were single parameters to differentiate malignant and benign lesions by univariate binary logistic regression analysis (P < 0.05).

Table 1

Conventional CT imaging characteristics of lesions.
Parameters	Benign(n = 67)	Malignant(n = 23)	Mann‒Whitney U test or χ² test	Univariate analysis Logistic regression analysis
Mean Size (mm) ²	26.06 ± 9.65	43.26 ± 21.16	P<0.001^#	P<0.001
<40	63(94%)	10(43.5%)
>40	4(6%)	13(56.5%)
Shape¹			P<0.001^*	P<0.001
Round	44(65.7%)	5(21.7%)
Lobulated	20(29.9%)	5(21.7%)
Irregular	3(4.4%)	13(56.5%)
Margin¹			P<0.001^*
Clear	67(100%)	10(43.5%)
Undefined	0(0%)	13(56.5%)
Enhancement pattern¹
Homogeneous	26(38.9%)	6(26.1%)
Heterogeneous	33(49.3%)	17(73.9%)
Rim	8(12.8%)	0(0%)
Calcification¹	1(1.5%)	4(17.4%)	P<0.05^*	P<0.05
Necrosis or Cysts¹	28(41.8%)	8(34.8%)
Lymphadenectasis¹	0(0%)	10(43.5%)	P<0.001^*
Invasion of adjacent tissues¹	1(1.5%)	10(43.5%)	P<0.001^*	P<0.001

^# Indicates a statistically significant difference according to the Mann‒Whitney U test

*Indicates a statistically significant difference according to χ² test (or Fisher’s exact test)

3.3 CT texture features and statistical analysis

The values of CT texture features and the results of Mann‒Whitney’s U test and univariate analysis logistic regression analysis are shown in Table 2. The CT texture analysis demonstrated that 7 histogram features, 10 GLCM features, and 11 GLRLM features showed apparent differences between benign and malignant lesions by Mann‒Whitney’s U test. Moreover, 5 histogram features, 9 GLCM features, and 3 GLRLM features showed significant differences between benign and malignant tumours by univariate analysis logistic regression analysis. Histogram parameters showed a trend of higher max intensity, volume count and kurtosis in malignant lesions than in benign lesions at the early phase and higher median intensity and mean value in the malignant group at the delayed phase. The energy, correlation, inverse difference moment, and Haralick correlation values, which are GLCM parameters, were higher in the malignant lesions than in the benign lesions in the early phase (P < 0.05). Conversely, Haralick correlation and the entropy values in the early phase were higher in benign lesions (P < 0.05 and P < 0.001, respectively). Moreover, entropy, cluster prominence, and Haralick correlation values in malignant lesions in the delayed phase were lower than those in benign lesions. GLRLM parameters showed that grey-level nonuniformity and run-length nonuniformity values in the early phase were greater in the malignant group. In addition, the grey level nonuniformity of malignant lesions in the delayed phase was more remarkable than that of benign lesions.

Table 2

The values of CT texture features and results of Mann‒Whitney’s U test analysis and univariate logistic regression analysis.
Parameters	Benign(n = 67)	Malignant(n = 23)	Mann‒Whitney U test	Univariate analysis Logistic regression analysis
Histogram Parameters
^eMax Intensity	112.2090 ± 49.1055	235.2609 ± 230.6444	P<0.05	P<0.05
^eVolume Count	22756.9851 ± 2564.8255	48665.2174 ± 46615.2992	P<0.05	P<0.05
^eKurtosis	4.1404 ± 2.7528	20.6445 ± 39.0384	P<0.001	P<0.05
^eVoxel Value Sum	1157205.46 ± 1500893.503	570712.48 ± 1672676.17	P<0.001
^dMedian Intensity	61.1403 ± 20.5507	72.9 ± 517.9413	P<0.05	P<0.05
^dMean Value	60.2166 ± 19.8532	72.3660 ± 18.3093	P<0.05	P<0.05
^dVoxel Value Sum	1068159.51 ± 1304974.28	243958.61 ± 359416.47	P<0.001
GLCM Parameters
^eEnergy	0.0471 ± 0.0222	0.1585 ± 0.1749	P<0.001	P<0.05
^eEntropy	5.1203 ± 0.5247	3.7305 ± 1.337	P<0.001	P<0.001
^eInertia	1.95 ± 1.10	1.98 ± 4.49	P<0.001
^eCorrelation	0.2067 ± 0.0830	0.5537 ± 0.6011	P<0.001	P<0.05
^eInverse Difference Moment	0.6109 ± 0.0712	0.6757 ± 0.2002	P<0.05	P<0.05
^eCluster Prominence	909.1647 ± 819.2049	359.3035 ± 685.8099	P<0.001	P<0.05
^eHaralick Correlation	11841.6720 ± 6524.2667	6048.8716 ± 10477.4282	P<0.001	P<0.05
^dEntropy	5.3255 ± 0.5745	4.2468 ± 1.8884	P<0.05	P<0.05
^dCluster Prominence	1056.7950 ± 793.0980	633.2993 ± 551.1701	P<0.05	P<0.05
^dHaralick Correlation	13892.8963 ± 7123.8264	10036.8509 ± 8549.8724	P<0.05	P<0.05
GLRLM Parameters
^eShortRunEmphasis	0.9999 ± 0.0002	0.0008 ± 0.0006	P<0.05
^eLongRunEmphasis	1.0006 ± 0.0095	1.0008 ± 0.0224	P<0.05
^eGreyLevelNonuniformity	447.2802 ± 500.3624	1022.2567 ± 917.9807	P<0.05	P<0.05
^eRunLengthNonuniformity	21513.4819 ± 24107.1684	46189.0996 ± 44620.5753	P<0.05	P<0.05
^eHighGreyLevelRunEmphasis	153396.4776 ± 55943.2891	115437.6613 ± 82703.5978	P<0.05
^eShortRunHighGreyLevelEmphasis	151411.2513 ± 51424.5131	114952.5169 ± 81058.5928	P<0.05
^eLongRunLowGreyLevelEmphasis	0.0094 ± 0.0140	0.015 ± 0.0406	P<0.05
^dShortRunEmphasis	0.9998 ± 0.0003	0.9999 ± 0.002	P<0.05
^dLongRunEmphasis	1.0007 ± 0.0013	1.0004 ± 0.0071	P<0.05
^dGreyLevelNonuniformity	352.7764 ± 416.8704	734.2822 ± 887.4893	P<0.05	P<0.05
^dLongRunLowGreyLevelEmphasis	0.0013 ± 0.0023	0.0081 ± 0.0012	P<0.05

^e: early phase, ^d: delayed phase. Data are the mean ± SD. P < 0.05 was considered to indicate a significant difference.

3.4 Multivariate binary logistic regression and ROC analysis

All independent factors screened by univariate analysis and multivariate analysis are shown in Table 3. Moreover, it was demonstrated that 3 models of predicted values were not significantly different from observed values by the Hosmer and Lemeshow test (P₁ = 0.334, P₂ = 0.633, P₃ = 0.104, P>0.05). All independent factors for each model are depicted in Table 4. For Model 1, the shape and invasion of adjacent tissues were independent factors for malignant lesions. For Model 2, entropy and inverse difference moment were independent factors for malignancy. The entropy, inverse difference moment and invasion of adjacent tissues were independent predictors for malignancy in Model 3. The ROCs of the 3 models are depicted in Fig. 2. The diagnostic abilities of these 3 models are demonstrated in Table 4. As can be seen from the above that the diagnostic accuracy of Model 3 was superior to those of the other two models. The Hanley and McNeill test was applied to compare the AUC of the 3 models, and the results indicated that Model 1 and Model 3 were significantly different (P_{1 vs 3} = 0.043 < 0.05), while there was no difference between Models 1 and 2 or Models 2 and 3 (P_{1 vs 2} = 0.152; P_{2 vs 3} = 0.183 > 0.05).

Table 3

ROC analyses of morphological parameters and CT texture features were significant for discriminating between malignant and benign salivary gland lesions.
Parameters	P value	B	OR(95%CI)	AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	Accuracy
Shape	0.006	1.276	3.581(1.466–8.132)	0.799(0.679–0.919)	0.435(0.239–0.651)	0.955(0.866–0.988)	0.769(0.460–0.938)	0.831(0.725–0.904)	0.856
Invasion of adjacent tissues	0.017	1.296	13.895(1.465-131.821)	0.579(0.435–0.724)	0.826(0.604–0.943)	0.985(0.908–0.999)	0.950(0.730–0.997)	0.943(0.853 − 0.724)	0.778
^eKurtosis	0.097	0.035	0.001(0-0.044)	0.754(0.622–0.887)	0.833(0.365–0.991)	0.786(0.680–0.865)	0.217(0.083–0.442)	0.985(0.909–0.999)	0.789
^eEntropy	0.002	1.984	0.933(0.873–0.997)	0.887(0.794–0.981)	0.565(0.349–0.761)	0.970 (0.887–0.995)	0.167(0.099–0.263)	0.867(0.764–0.931)	0.867
^eInverseDifferenceMoment	0.007	12.966	<0.001	0.759(0.632–0.887)	0.304(0.141–0.530)	0.970 (0.887–0.995)	0.778(0.402–0.961)	0.802(0.696–0.880)	0.8

e: early phase.

Table 4

ROC analyses of morphological parameters and CT texture features were significant for discriminating between malignant and benign salivary gland lesions.
Parameters	AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	Accuracy
Model 1	0.818(0.698–0.938)	0.652(0.428–0.828)	0.955(0.866–0.988)	0.833(0.577–0.956)	0.889(0.787–0.947)	0.878
Model 2	0.918(0.853–0.982)	0.696(0.470–0.859)	0.97(0.887–0.995)	0.889(0.639–0.981)	0.903(0.804–0.957)	0.9
Model 3	0.948(0.891-1.000)	0.783(0.558–0.917)	0.985(0.909–0.999)	0.947(0.718–0.997)	0.930(0.837–0.974)	0.933

Previous studies[17–19] have shown that two-phase CT is helpful for identifying various benign and malignant neoplasms in the parotid gland according to the CT enhanced pattern. Obviously, two-phase MSCT to access neoplasms only by CT enhanced pattern is not comprehensive. Texture analysis, which can characterize tissue organization by studying the spatial variation in the intensity of image pixel values, has greatly enriched information on pathologic features. Based on two-phase MSCT texture analysis of statistical parameters and MSCT image features, our study results demonstrated that two-phase MSCT texture analysis was conducive to differentiating between malignant and benign neoplasms in the salivary gland, especially when combined with morphological CT image features.

We statistically analysed these parameters of the texture analysis. First, the results revealed that the early phase had the advantage of differentiating between malignant and benign neoplasms in the salivary gland, based on the SGTs blood supply and histopathological characteristics. Moreover, compared with the benign group, malignant lesions had a lower entropy value in the early phase. In other words, malignant lesions were more random and dispersed due to the heterogeneity of the neoplasm. Our findings were similar to a CT texture analysis’s work focused on radiation-induced changes in parotids[20, 21]. The authors found a obvious decrease in local entropy in postradiation therapy subjects due to the reduction of acinar cells in the parotids. Our results can be interpreted in a similar way: normal acinar cells of the salivary gland were replaced with tumour cells in an disorganized way, which resulted in increasing randomness of the lesion area.

In addition, we also found that the kurtosis value of malignant tumours was higher than that of benign tumours in the early phase. Previous research[22] suggested that the kurtosis value was increased by intensity variations in highlighted objects (i.e., blood vessels). Scrima et al. found that kurtosis was associated with neovascularity in small renal masses[23]. Another study indicated that a higher rate of angiogenesis was noted in malignant tumours than in benign tumours by immunohistochemistry[24, 25]. Accordingly, we hypothesized that intratumoral microvessel density was elevated more obviously in malignant neoplasms, which was positively associated with the kurtosis value. Nevertheless, from our results, kurtosis is not an independent factor for differentiating between malignant and benign neoplasms. The reason could be that both malignant tumours and Warthin tumours tend to have higher microvascularity, which might cause no significant difference between them.

Finally, our results showed that the inverse difference moment value, inversely associated with local spatial variation, was higher in the malignant lesions than in the benign lesions. This means that local spatial variation in the benign salivary glands was more obvious. Previous research has suggested that cystic degeneration and tumour necrosis are common imaging features in SGTs, including pleomorphic adenomas[26], Warthin tumours[27], basal cell adenomas[28, 29], adenoid cystic carcinomas[30], mucoepidermoid carcinomas[30], acinic cell carcinomas[31], and salivary duct carcinomas[32]. One reason may be that cystic or necrotic regions in benign tumours were more common than malignant lesions in our results, which would cause higher local spatial variation in image texture.

Clinically, the size, morphology, vessels of lesions and nodal metastases were analysed on conventional CT images to differentiate between malignant and benign neoplasms. We used statistical methods to construct the 3 models by morphological parameters, MSCT texture analysis parameters and a combination of both. Our results demonstrated that the diagnostic capability (diagnostic accuracy, sensitivity and specificity) of morphological CT parameters was lower than that of MSCT texture analysis parameters, including kurtosis, entropy and inverse difference moment. (Model 1 87.8%, 65.2%, 95.5%; Model 2 90%, 69.6%, 97%, respectively). Furthermore, Model 3 constructed by the morphological CT features and MSCT texture analysis parameters had the highest diagnostic accuracy, sensitivity, and specificity among Models 1 and 2 (93.3%, 78.3%, 98.5%, respectively). Notably, Model 3 had greatly improved sensitivity from 65.2–78.3% compared with Model 1.

As seen from the above, 2-phase MSCT texture analysis parameters, especially combined with morphological CT features, can enhance diagnostic accuracy and help us gain insight into the underlying biological characteristics of salivary lesions.

the limitations of the study are presented as bellow. Firstly, our study was a retrospective single-centre analysis, and the inclusion of only 90 patients in a few years was not enough. Further long-term and multicentre studies are needed to confirm the results in larger cohorts. Next, the population of our study only included several types of malignant and benign salivary neoplasms, and more pathological types of tumours should be included in further research. Finally, even if 2-phase MSCT texture analysis can help differentiate malignant salivary tumours, it is still not adequate enough to analyse the complicated pathological features of salivary gland tumours. More research is needed to identify a direct correlation between pathology and CT texture analysis parameters.

Two-phase MSCT texture analysis was conducive to differentiating between malignant and benign neoplasms in the salivary gland, especially when combined with morphological CT features. MSCT texture analysis can be used to enrich information obtained from CT in the differentiation between malignant and benign neoplasms and guide clinical staging and therapy of neoplasms in terms of underlying biological characteristics. However, this technique has not yet been routinely applied in clinical practice. Further study is needed to confirm the results in larger samples and make a direct correlation between pathology and CT texture analysis parameters.

MSCT: multislice spiral computed tomography; ROC: receiver operating characteristics; SGTs: Salivary gland tumours; GLCM: grey-level cooccurrence matrix parameters; GLRLM: grey-level run length maxia parameters; ROC: receiver operating characteristic; AUC: area under the receiver operatingcurve; PPV: positive predictive value; NPV: negative predictive value; OR: odds ratio; CI: confdence interval. e: early phase, d: delayed phase.

Ethics approval and consent to participate

This study was authorized by Ethics Committee of the First Afﬁliated Hospital of Guangzhou University of Chinese Medicine (Guangzhou University of Chinese Medicine, Guangzhou, China), and patient informed consent was waived for this type of review.

Consent for publication

Informed consent was obtained from all patients.

Data Availability Statement

The datasets generated and analysed during the current study are not publicly available due protecting privacy of the patients, but are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interests.

Funding

Project supported by the Natural Science Foundation of Guangdong Province, China (Grant No. 2022A1515011028).

Author Contributions

Conceptualization, J.Z. and S.Q.; methodology and investigation, S.J.; Y.S.,Y.L.;Z.Z.; M.L.; formal analysis, S.J.; Y.S.,Y.L.; writing—original draft preparation, S.J.; data curation, S.J.; Y.S. and Y.L.; writing—review and editing, Y.S., J.Z. and S.Q.; project administration, J.Z.; funding acquisition, J.Z. All authors have read and agreed to the published version of the manuscript.

Acknowledgements

Not applicable.

Sumi M, Van Cauteren M, Sumi T, Obara M, Ichikawa Y, Nakamura T: Salivary gland tumors: use of intravoxel incoherent motion MR imaging for assessment of diffusion and perfusion for the differentiation of benign from malignant tumors. Radiology 2012, 263(3):770–777.
Pinkston JA, Cole P: Incidence rates of salivary gland tumors: results from a population-based study. Otolaryngology–head and neck surgery: official journal of American Academy of Otolaryngology-Head and Neck Surgery 1999, 120(6):834–840.
Vogl TJ, Albrecht MH, Nour-Eldin NA, Ackermann H, Maataoui A, Stöver T, Bickford MW, Stark-Paulsen T: Assessment of salivary gland tumors using MRI and CT: impact of experience on diagnostic accuracy. La Radiologia medica 2018, 123(2):105–116.
Choi SY, Lee E, Kim E, Chung MK, Son YI, Baek CH, Jeong HS: Clinical outcomes of bulky parotid gland cancers: need for self-examination and screening program for early diagnosis of parotid tumors. 2021, 21(1):178.
Fakhry N, Aldosari B, Michel J, Giorgi R, Collet C, Santini L, Giovanni A, Dessi P: Clinical and oncological outcomes after surgical excision of parotid gland tumours in patients aged over 80 years. International journal of oral and maxillofacial surgery 2013, 42(11):1385–1390.
Xu B, Saliba M, Ho A, Viswanathan K, Alzumaili B, Dogan S, Ghossein R, Katabi N: Head and Neck Acinic Cell Carcinoma: A New Grading System Proposal and Diagnostic Utility of NR4A3 Immunohistochemistry. The American journal of surgical pathology 2022, 46(7):933–941.
Vogl TJ, Albrecht MH, Nour-Eldin NA, Ackermann H, Maataoui A, Stover T, Bickford MW, Stark-Paulsen T: Assessment of salivary gland tumors using MRI and CT: impact of experience on diagnostic accuracy. La Radiologia medica 2018, 123(2):105–116.
Lee YY, Wong KT, King AD, Ahuja AT: Imaging of salivary gland tumours. European journal of radiology 2008, 66(3):419–436.
Ogawa M, Osaga S, Shiraki N, Kawakita D, Hanai N, Tamaki T, Tsukahara S, Kawaguchi T, Urano M, Shibamoto Y: Utility of CT texture analysis to differentiate olfactory neuroblastoma from sinonasal squamous cell carcinoma. Scientific reports 2021, 11(1):4679.
Ahn SJ, Kim JH: CT reconstruction algorithms affect histogram and texture analysis: evidence for liver parenchyma, focal solid liver lesions, and renal cysts. 2019, 29(8):4008–4015.
Owen B, Gandara D, Kelly K, Moore E, Shelton D, Knollmann F: CT Volumetry and Basic Texture Analysis as Surrogate Markers in Advanced Non-small-cell Lung Cancer. Clinical lung cancer 2020, 21(3):225–231.
Yeung J, Ganeshan B: Equilibrium CT Texture Analysis for the Evaluation of Hepatic Fibrosis: Preliminary Evaluation against Histopathology and Extracellular Volume Fraction. 2020, 10(2):46.
Humphries SM, Yagihashi K, Huckleberry J, Rho BH, Schroeder JD: Idiopathic Pulmonary Fibrosis: Data-driven Textural Analysis of Extent of Fibrosis at Baseline and 15-Month Follow-up. 2017, 285(1):270–278.
Yang G, Gong A, Nie P, Yan L, Miao W, Zhao Y, Wu J, Cui J, Jia Y, Wang Z: Contrast-Enhanced CT Texture Analysis for Distinguishing Fat-Poor Renal Angiomyolipoma From Chromophobe Renal Cell Carcinoma. 2019, 18:1536012119883161.
Chen C, Ou X, Li H, Zhao Y, Zhao F, Zhou S, Ma X: Contrast-Enhanced CT Texture Analysis: a New Set of Predictive Factors for Small Cell Lung Cancer. 2020, 22(3):745–751.
Crimì F, Zanon C, Cabrelle G: Contrast-Enhanced CT Texture Analysis in Colon Cancer: Correlation with Genetic Markers. 2022, 8(5):2193–2201.
Choi DS, Na DG, Byun HS, Ko YH, Kim CK, Cho JM, Lee HK: Salivary gland tumors: evaluation with two-phase helical CT. Radiology 2000, 214(1):231–236.
Joo YH, Kim JP, Park JJ, Woo SH: Two-phase helical computed tomography study of salivary gland warthin tumors: a radiologic findings and surgical applications. Clinical and experimental otorhinolaryngology 2014, 7(3):216–221.
Woo SH, Choi DS, Kim JP, Park JJ, Joo YH, Chung PS, Kim BY, Ko YH, Jeong HS, Kim HJ: Two-phase computed tomography study of warthin tumor of parotid gland: differentiation from other parotid gland tumors and its pathologic explanation. Journal of computer assisted tomography 2013, 37(4):518–524.
Scalco E, Fiorino C, Cattaneo GM, Sanguineti G, Rizzo G: Texture analysis for the assessment of structural changes in parotid glands induced by radiotherapy. Radiotherapy and oncology: journal of the European Society for Therapeutic Radiology and Oncology 2013, 109(3):384–387.
Nardone V, Tini P, Nioche C, Mazzei MA, Carfagno T, Battaglia G, Pastina P, Grassi R, Sebaste L, Pirtoli L: Texture analysis as a predictor of radiation-induced xerostomia in head and neck patients undergoing IMRT. La Radiologia medica 2018, 123(6):415–423.
Miles KA, Ganeshan B, Hayball MP: CT texture analysis using the filtration-histogram method: what do the measurements mean? Cancer imaging: the official publication of the International Cancer Imaging Society 2013, 13(3):400–406.
Scrima AT, Lubner MG, Abel EJ, Havighurst TC, Shapiro DD, Huang W, Pickhardt PJ: Texture analysis of small renal cell carcinomas at MDCT for predicting relevant histologic and protein biomarkers. Abdominal radiology (New York) 2019, 44(6):1999–2008.
Theocharis S, Gribilas G, Giaginis C, Patsouris E, Klijanienko J: Angiogenesis in salivary gland tumors: from clinical significance to treatment. Expert opinion on therapeutic targets 2015, 19(6):807–819.
Tadbir AA, Pardis S, Ashkavandi ZJ, Najvani AD, Ashraf MJ, Taheri A, Zadeh MA, Sardari Y: Expression of Ki67 and CD105 as proliferation and angiogenesis markers in salivary gland tumors. Asian Pacific journal of cancer prevention: APJCP 2012, 13(10):5155–5159.
Kato H, Kanematsu M, Mizuta K, Aoki M: Imaging findings of parapharyngeal space pleomorphic adenoma in comparison with parotid gland pleomorphic adenoma. Japanese journal of radiology 2013, 31(11):724–730.
Ikeda M, Motoori K, Hanazawa T, Nagai Y, Yamamoto S, Ueda T, Funatsu H, Ito H: Warthin tumor of the parotid gland: diagnostic value of MR imaging with histopathologic correlation. AJNR American journal of neuroradiology 2004, 25(7):1256–1262.
Chawla AJ, Tan TY, Tan GJ: Basal cell adenomas of the parotid gland: CT scan features. European journal of radiology 2006, 58(2):260–265.
Okahara M, Kiyosue H, Matsumoto S, Hori Y, Tanoue S, Uchida D, Mori H, Kondo Y: Basal cell adenoma of the parotid gland: MR imaging findings with pathologic correlation. AJNR American journal of neuroradiology 2006, 27(3):700–704.
Okahara M, Kiyosue H, Hori Y, Matsumoto A, Mori H, Yokoyama S: Parotid tumors: MR imaging with pathological correlation. European radiology 2003, 13 Suppl 4:L25-33.
Suh SI, Seol HY, Kim TK, Lee NJ, Kim JH, Kim KA, Woo JS, Lee JH: Acinic cell carcinoma of the head and neck: radiologic-pathologic correlation. Journal of computer assisted tomography 2005, 29(1):121–126.
Weon YC, Park SW, Kim HJ, Jeong HS, Ko YH, Park IS, Kim ST, Baek CH, Son YI: Salivary duct carcinomas: clinical and CT and MR imaging features in 20 patients. Neuroradiology 2012, 54(6):631–640.

No competing interests reported.

SupplementaryMaterial.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Imaging prediction of malignant and benign neoplasms in the salivary gland using two-phase multislice spiral CT texture analysis

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

1. Background

2. Methods

2.1. Study Population

2.2. CT imaging protocol

2.3. CT qualitative image analysis

2.4. CT texture analysis

2.5. Statistical Analysis

3. Results

3.1 Study population

3.2 Characteristics of lesions and statistical analysis

3.3 CT texture features and statistical analysis

3.4 Multivariate binary logistic regression and ROC analysis

4. Discussion

5. Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1