1.1 Patients and MRI acquisition parameters
Patients
Eight patients were retrospectively reviewed (five men, three women; mean age was 51 [range 13–88] years) with a biopsy-proven malignant mesenchymal tumour. The sample patients had preoperative MRI scans between May 2018 and June 2020, and the pathological differentiation could be determined. The sample included three sarcomas and five lymphomas, including one low-grade central osteosarcoma of the left zygoma, two well-differentiated chondrosarcomas of the jaw, one well-differentiated diffuse large B-cell tumour of the buccal region, one B-cell lymphoma related marginal area of tongue mucosa and three non-Hodgkin's follicular lymphoma of the parotid gland. Only patients with sarcomas showed varying degrees of pain and limited facet joint movement, which had no specific clinical symptoms.
The general characteristics of the participants were shown in Table 1. The inclusion criteria: biopsy-proven malignant mesenchymal tumour without concomitant disease. The exclusion criteria were as follows: without definitive post-operative information on pathological characteristics, a minimum tumour diameter < 5 mm, poor MRI quality.
Table 1
The clinical and MRI characteristics of patients (n = 8)
MR1 scan
|
Sex
|
Age
|
Region
|
ADC valuc(l0 3mm2/s)
|
Pathological and IHC
|
Patient 1
|
F
|
26
|
zygoma
|
1.12
|
spindie shaped tumor cells and scattered in trabecular bone tissue and bone like matrix tissue; Ki-67( 10–20%), CK(-), SMA and CD99 (+)
|
Patient 2
|
F
|
40
|
jaw
|
1.56
|
a large number of chondrocytes with obvious heteromorphism and bone septum; viaentin (+), S-100 and CK (-), Ki-67(60%)
|
Patient 3
|
M
|
56
|
jaw
|
1.54
|
A large number of chondrocytes; S-100(-), Viacntin(+), CK(-), Ki-67(50%)
|
Patient 4
|
F
|
62
|
buccal
|
0.56
|
Lymphoid hyperplasia; PCK(-), EMA(-), CD20(+), CD79a(+), PAX-5(+), CD3(-), CD38(+), CyclinDI(-), MUMl(-), CD30(-), CD 10(+), Bcl-6(-), Bcl-2(-), CD23(-), CD5(+), Ki67(70%)
|
Patient 5
|
M
|
88
|
tongue
|
0.32
|
Lymphoid hyperplasia, destroyed lymphoid follicles structure; CD3T(+), BCL-6(-), BCL-2(+), CD 10(-), cyclinDl(-), CD79a(+), Pan-5(+), kappa(-), Ki-67(< 10%)
|
Patient 6
|
M
|
56
|
parotid gland
|
0.37
|
Lymphoid hyperplasia with obvious heteromorphism; CD3(+), CD20(+), BCL-2(+), BCL-6(+), CD 10(+), Muml(+), PAX-5(+), CD79a(+), Ki-67(70–80%)
|
Patient 7
|
M
|
67
|
parotid gland
|
0.54
|
Lymphoid hyperplasia, tumor cells infiltrated glands in some areas, serous acini and adipose tissue display; CD20(+), CD 10 (+), CD3 partial cells (+), CD21 showed FDC network, bcl-6 (+), bcl-2 (+), CD38 germinal center positive (+)
|
Patient 8
|
M
|
13
|
parotid gland
|
0.57
|
-ymphoid hyperplasia, CD3T(+), CD10(-), BCL-6(+), BCL-2(+), CD79a(+), Ki-67(40–50%)
|
Table 2
First-order features of sarcomas: ADCkurt、ADCskew、ADCmean and ADCmedian.
First-order features
|
Patient 1
|
Patient 2
|
Patient 3
|
Mean
|
ADCkurt
|
3.520680603
|
3.946304522
|
3.280937825
|
3.5826
|
ADCskew
|
0.627515252
|
0.350436023
|
0.391892136
|
0.4566
|
ADCmean
|
180.4899019
|
41.69653962
|
-95.37971946
|
42.2689
|
ADCmedian
|
191.38255039999999
|
39.44982554
|
-102.6499184
|
42.7275
|
Table 3
First-order features of lymphomas: ADCkurt, ADCskew, ADCmean and ADCmedian.
First-order features
|
Patient 4
|
Patient 5
|
Patient 6
|
Patient 7
|
Patient 8
|
Mean
|
ADCkrut
|
8.259956138
|
6.073467803
|
5.280964199
|
5.827922475
|
4.898098651
|
6.0681
|
ADCskew
|
2.142312781
|
1.741374425
|
1.5906607
|
1.302931488
|
1.369502429
|
1.6294
|
ADCmean
|
-63.95187937
|
-34.23833038
|
-103.367401
|
-39.59149499
|
-65.52257619
|
-61.3343
|
ADCmedian
|
-75.86062501
|
-48.1006346
|
-112.4609443
|
-44.98317419
|
-69.76222731
|
-70.2335
|
MRI Acquisition Parameters
1.5-T Siemens Avanto with an eight-channel phased-array neck coil was used in this study. The patient's head was secured. Non-contrast axial, sagittal and coronal FS-T2WI sequences acquired in multiple breath-holds were obtained by the following parameters: a repetition/echo time of 5080/87 ms, a slice thickness/interslice gap of 4.0/0.4 mm, 20 slices and a matrix of 256 × 320. Axial T1-weighted images were also acquired in multiple breath-holds. Diffusion-weighted images were obtained in the coronal plane. Following the image acquisition, a pixel-wise ADC map was generated by the inbuilt software using b values of 800 s/mm2. All patients received a 15-ml intravenous bolus injection of gadodiamide (GE Healthcare Ireland Limited, County Cork, Republic of Ireland). The contrast imaging was performed using a fat-suppressed three-dimensional (3D) T1-weighted volumetric interpolated breath-hold examination sequence after the injection.
The shape, size, signal, bone destruction, adjacent tissue relationship on MRI were evaluated. Besides, the ADC map was generated based on DWI, and the sampling was selected to measure the ADC value at the maximum level of the lesion. The lesions were resected surgically in all eight patients. Histopathological and immunohistochemical staining (IHC) was performed postoperatively.
1.2 MRI And Radiomics Analysis
Dr Wise Multimodal Research Platform was used for radiomics analysis. An open-source python package called PyRadiomics (2.2.0) was used for extraction of features. The platform supports feature extraction used to calculate single values per feature for an ROI (‘segment-based’) or generate feature maps (‘voxel-based’) (Fig. 1).
Delineation Of Tumour Roi
The tumour regions in the primary dataset were labeled manually by two experts. In the case of disagreement, a third opinion was requested. The DWI-ADC parameter diagram scan was selected as the labeling image, then tumour tissue was classified.
Extracting Features From Mri Scans
A B-spline interpolation resampling was used and the anisotropic voxels were resampled to form isotropic voxels of 2.0 mm × 2.0 mm × 2.0 mm. The MRI images were then normalised by centring it at the mean with standard deviation.
(s = 100; µᵡ represents mean value; σ represents standard deviation)
Eighteen first-order features were obtained from the original images based on the pixel value extracted from each ROI: Energy, TotalEnergy, MeanAbsolute Deviation, RobustMeanAbsolute Deviation, Entropy, 10Percentile, 90Percentile, Minimum, Maximum, Mean, Median, InterquartileRange, Range, RootMeanSquared, Skewness, Kurtosis, Variance and Uniformity.
1.3 Statistical Analysis
The study's group of data obeys the normal distribution and has the homogeneity of variance. SPSS 16.0 (IBM Corp., Armonk, NY, USA) was used. Group differences in quantitative variables were analzed by t-test. A P-value < 0.05 was considered statistically significant.