Purpose

The incidence of Cervical cancer (CC) ranks as the sixth most common female cancers in China [1]. The most common histopathological subtype of CC is squamous cell carcinoma (SCC) and adenocarcinoma (AC). CC exhibits a variety of heterogeneity at histopathological and genetic levels [2]. AC has an increased likelihood of metastasis and worse prognosis than SCC [3].It is very difficult to identify different subtypes with noninvasive imaging methods, and this is still a major clinical challenge. Many studies have shown that diffusion weighted imaging (DWI)/apparent diffusion coefficient (ADC) maps and dynamic contrast-enhanced (DCE) MRI can play important roles in differentiating different pathological subtypes of CC [4]. However, some authors have argued that there are no statistically significant differences in ADC values between SCC and AC [5]. Currently, radiomics is becoming a quantitative method to reveal and describe intratumoural heterogeneity [6,7]. The purpose of this study is to investigate whether MR radiomics features can differentiate AC and SCC of CC and to evaluate the differentiation capabilities of different MR series and radiomics features.

Methods

One hundred and forty-eight women with CC underwent pelvic MRI on a 3.0-T MR scanner (Siemens Healthcare, Erlangen, Germany) with a 32-channel phase-array body coil were retrospectively analysed. MRI protocol for radiomics analysis included T2WI turbo spin-echo and CE volumetric interpolated breath-hold sequences in 2 imaging planes (axial/sagittal) (T2TRA, T2SAG, CETRA, and CESAG), and axial echo planar imaging DWI sequences. The DWI sequences were acquired with one of two b value sets (b=50, 800 sec/mm2) and ADC maps were automatically produced. The parameters for radiomics were as follows: (1) T2TRA: TR/TE = 6500 or 6080/107 ms; section thickness (ST) = 7 mm; intersection gap = 1.4 mm; FOV = 262×350 or 281×360 mm; matrix = 360×640 or 225×384. (2) T2SAG: TR/TE = 3390 or 3600/77 ms; ST = 5 mm; intersection gap = 1 mm; FOV = 250×250 or 281×281 mm; matrix = 240×320. (3) ADC: TR/TE = 5500/87 ms; ST = 7 mm; intersection gap = 1.4 mm; FOV = 280×350 or 329×350 mm; matrix = 102×170 . (4) CESAG: TR/TE = 3.69/1.46 ms; flip angle (FA) = 9°; ST = 3 or 4 mm; intersection gap = 0 mm; FOV = 237×269 or 260×309 mm; matrix = 310×352 or 324×384. (5) CETRA: TR/TE = 3.57/1.39 ms; flip angle (FA) = 9°; ST = 4.5 mm; intersection gap = 0 mm; FOV = 349×349 or 378×378 mm; matrix = 384×384. The inclusion criteria were as follows: 1) MRI including DWI/ADC maps and contrast-enhanced (CE) MRI (< 2 months) before surgery and 2) no previous treatment before MRI. Patients with 1) a largest tumour diameter on MRI <1.0 cm (n=20) and 2) imaging artefacts on MRI on subjective assessment (n=32) were excluded from the analysis. Thus, the final population consisted of 96 CC patients (AC = 46; SCC = 50). Radiomics features were computed from 5 different original MR series. Significantly different radiomics features between AC and SCC were evaluated by the Wilcoxon rank-sum test. Clustering and logistic regression were used to examine the capabilities of (1) different radiomics features to classify AC and SCC and (2) 5 MR series to differentiate AC from SCC.

Results

Among the 105 extracted radiomics features, 51, 38, 37 and 2 features showed significant differences between AC and SCC on T2SAG, T2TRA, ADC and CESAG sequences, respectively (p<0.05). ACs had greater textural heterogeneity than SCCs. By unsupervised clustering of significantly different features, T2SAG sequences achieved the highest accuracy of 0.833 (sensitivity: 0.920; specificity: 0.739) for differentiating SCC from AC, followed by ADC (accuracy: 0.75; sensitivity: 0.780; specificity: 0.717) and T2TRA (accuracy 0.719; sensitivity: 0.880; specificity: 0.543) sequences. The best area under the curve (AUC) for classification ability was 0.86 for T2SAG sequences (accuracy: 0.81; sensitivity: 0.74; specificity: 0.88), followed by ADC (AUC: 0.82; accuracy: 0.75; sensitivity: 0.67; specificity: 0.82) and T2TRA (AUC: 0.79; accuracy: 0.75; sensitivity: 0.63; specificity: 0.86) sequences (Fig.1).

Discussion and conclusion

In conclusion, we found that the use of radiomics features derived from multiple MR series was a feasible noninvasive method for differentiating between AC and SCC. AC had more heterogeneous radiomics features than SCC. Radiomics features on routine T2WI and ADC maps can supply additional information to classify subtypes of CC, and T2SAG was the best imaging sequence.

Acknowledgements

No acknowledgement found.

References

1 Chen W, Zheng R, Baade PD et al (2016) Cancer statistics in China, 2015. CA Cancer J Clin 66:115-132 2 Guan Y, Li W, Jiang Z et al (2016) Whole-Lesion Apparent Diffusion Coefficient-Based Entropy-Related Parameters for Characterizing Cervical Cancers: Initial Findings. Acad Radiol 23:1559-1567 3 Kuang F, Ren J, Zhong Q, Liyuan F, Huan Y, Chen Z (2013) The value of apparent diffusion coefficient in the assessment of cervical cancer. Eur Radiol 23:1050-1058 4 Kuang F, Yan Z, Li H, Feng H (2015) Diagnostic accuracy of diffusion-weighted MRI for differentiation of cervical cancer and benign cervical lesions at 3.0T: Comparison with routine MRI and dynamic contrast-enhanced MRI. Journal of Magnetic Resonance Imaging 42:1094-1099 5 Downey K, Riches SF, Morgan VA et al (2013) Relationship between imaging biomarkers of stage I cervical cancer and poor-prognosis histologic features: quantitative histogram analysis of diffusion-weighted MR images. AJR Am J Roentgenol 200:314-320 6 Gillies RJ, Kinahan PE, Hricak H (2016) Radiomics: Images Are More than Pictures, They Are Data. Radiology 278:563-577 7 Avanzo M, Stancanello J, El Naqa I (2017) Beyond imaging: The promise of radiomics. Phys Med 38:122-139