Quantitative texture analysis based on dynamic contrast MRI for the differential diagnosis between primary thymic lymphoma and thymic carcinoma

The patients

This retrospective study was approved by the Institutional Review Board of People’s Hospital of Jiangsu Province and the First Affiliated Hospital of Nanjing Medical University, and the requirement to obtain informed consent from the patient was lifting (permit number: 2021-SR-238). All methods were performed in accordance with current guidelines and regulations.

In this retrospective study, we reviewed the medical records of patients with thymic carcinoma and thymic lymphoma at our hospital from April 2018 to March 2021. Patients who met the following criteria were included: (1) primary tumors all were confirmed by surgery or biopsy by percutaneous puncture; (2) routine MRI and DCE-MRI parameters were complete; (3) no operation, puncture, radiotherapy or hormone therapy was performed before the MRI examination. We excluded 12 patients for the following reasons: (1) inadequate MRI quality (n = 4); (2) treated before the examination (n = 8). Finally, we included a total of 68 pathologically diagnosed patients, including 32 patients with thymic carcinoma (22 men, 10 women, mean age 55.4 ± 13.1 years) and 37 patients with thymic lymphoma (18 men, 19 women, mean age 36.4 ± 14.9 years). years).

Imaging protocol

All MRI examinations were performed using a 3 T MRI system (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany) with a 16-channel chest coil. All patients underwent conventional MRI and MRI-DCE from the suprasternal notch to the diaphragm. Axial DCE-MRI used the StarVIBE sequence which allowed the patient to breathe freely. Conventional imaging protocols included unenhanced axial T1-weighted imaging (repetition time (TR) of 140 ms, echo time (TE) of 2.5 ms) and coronal T2-weighted imaging (TR of 1200 ms, TE of 93 ms). A bolus of gadolinium-diethylene triamine pentaacetic acid (Magnevist; Bayer Schering Pharma AG, Berlin, Germany) was injected into the elbow vein via an electric injector with a flow rate of 4.0 mL/s at a dose of 0. 1 mmol/kg, followed by a bolus of 20 ml saline administered at the same injection rate. During the entire acquisition process, First, three unenhanced datasets were acquired using T1W imaging starVIBE with flip angles of 5°, 10°, and 15°, respectively, to obtain the map T1. Second, the dynamic sequence was acquired after the baseline T0 acquisitions and thirty-one sets of contrast-enhanced images were acquired. StarVIBE DCE-MRI detailed imaging parameters were: 3.19ms TR/1.13ms TE, 3mm slice thickness, 400mm2 field of view (FOV), 160*224 matrix, 15° flip angle. The temporal resolution was 8.8 s and the total acquisition time was 5 min 8 s.

Image processing

DCE-MRI data were downloaded and processed with in-house software (Omin-Kinetics; GE Healthcare, Shang Hai, People’s Republic of China). For arterial input function (AIF) selection, a freehand region of interest (ROI) was placed in the descending aorta on the DCE-MRI images. Average ROI size ranged from 6 to 9 mm2. The AIF curve has been approved by a senior chest radiologist to ensure its accuracy. The Extended Tofts Linear two-compartment model was used to calculate pharmacokinetic parameters. Determine the location of the lesion by combining T2WI and DCE, adjust the image to the phase with the most obvious enhancement, manually draw the ROI on each cross-section, and merge the lesions in the software to generate the three-dimensional (3D) ROI -KING ). The measurement was made along the edge of the lesion tissue, ensuring that the return on investment was lower than that of the lesion, reducing the effect of the volume effect, and making the lesion tissue in the region of more representative interest. Necrotic, cystic and hemorrhagic areas should be avoided as much as possible. Parametric maps derived from DCE, including volume transport constant (Ktrans), plasma rate constant (Kep), and the percentage of extracellular space volume (Ve) were calculated automatically based on the Tofts model. Texture parameters were acquired using the same software (Omin-Kinetics; GE Healthcare, Shang Hai, PR China). Features used in our study include mean, median, 5th/95th percentile (P5/P95), skewness, kurtosis, diff-variance, diff-entropy, contrast, and entropy.

Texture analysis of DCE-MRI images was performed by two senior thoracic radiologists with 7 and 3 years of experience, both blinded to clinical information and final histopathological findings. Measurements from both readers were used for assessment of interobserver reproducibility.

statistical analyzes

All statistical analyzes were performed using SPSS (version 26.0, Chicago, IL, USA) and MedCalc (version 20.0.4, Mariakierke, Belgium) software package. The normality of the data distributions was analyzed using the Kolmogorov-Smirnov test. All numerical data with normal distributions were reported as mean ± standard deviation. Otherwise, medians (25th–75th percentiles) were reported. Independent sample t-test or Mann-Whitney U-test was used to compare differences in texture parameters between the two groups. Logistic regression was used to select the parameters and the receiver operating characteristic curve (ROC) was used to assess the diagnostic value of each parameter to differentiate thymic carcinoma and thymic lymphoma. P

The inter-observer reproducibility of parameter measurement in this study was assessed using the intraclass correlation coefficient (ICC) with 95% confidence intervals (CI) and applying a two-way ICC with an assumption of random evaluator. The ICC has been interpreted as follows: 0.81, excellent.

Previous These startups want to make credit scores a thing of the past – TechCrunch
Next 1 action on energy, 1 on gold and 1 on copper for the second half of 2022