Introduction/purpose

Sinonasal mucosal melanoma (SNMM) is a rare oncological entity which requires precise diagnosis to guide timely surgery.(1) However, preoperative biopsy is a challenge since it could not provide intact information nor avoid metastasis during the puncture. Besides, due to its rarity, precise differentiation of SNMM depending on visible characteristics on MRI from sinonasal lymphoma is also difficult for radiologists.(2) As a sound and noninvasive method, radiomics enables reflecting intra-tumor histopathological properties through extracting quantitative features unperceived by naked eyes. The objective of the current retrospective study was to determine whether MRI-based radiomics could be helpful in the preoperative discrimination of rare SNMM from sinonasal lymphoma.

Methods

This retrospective study involved 93 patients with sinonasal malignancies including 48 SNMM and 45 sinonasal lymphoma. T2-weighted images (T2WI) and post-contrast T1-weighted images (C-T1WI) were acquired on a 3T MR scanner (MAGNETOM Verio, Siemens Healthcare, Erlangen, Germany), and a 32-channel head coil was used. A total number of 214 quantitative features was extracted respectively from the volume of interest (VOI) or representative region of interest (rROI) of the tumor through an open-resource tool FAE (https://github.com/salan668/FAE) based on Python 3.5, and 3 quality features (age, gender, and involved or uninvolved) were also evaluated. According to the generated 217 features, an individualized 3D and an individualized 2D nomogram were built from the VOI and the rROI, respectively. Further, based on the 214 quantitative features, a generalized 3D nomogram was calculated from the VOIs. Meanwhile, the results by two radiologists (3 years’ and 20 years’ experience in MRI imaging, respectively), who only had access to T2WI and C-T1WI, were obtained. The differentiation accuracy between the radiomics-based individualized 3D nomogram and the visual assessment by radiologist were compared, as well as the area under the curve (AUC) between the individualized 3D nomogram and individualized 2D nomogram, or the AUC between the individualized 3D nomogram and the generalized 3D nomogram were further conducted.

Results

The performance of the individualized 3D nomogram to distinguish SNMM from sinonasal lymphoma was good with an accuracy of 0.88, AUC of 0.94 (95% confidence interval [CI]: 0.91 to 0.96) in the training cohort and an accuracy of 0.89, AUC of 0.93 (95% CI: 0.87 to 0.98) in the validation cohort (Fig 1). Compared with visual assessments of both the new radiologist (p<0.001) and the experienced radiologist (p<0.05), the individualized 3D nomogram demonstrated a higher accuracy (Fig 2). In addition, with a slightly higher AUC (all p<0.001), the individualized 3D nomogram showed higher diagnosis value than the individualized 2D nomogram (Fig 3) and the generalized 3D nomogram (Fig 4) based on only quantitative radiomics features.

Discussion

By employing large amounts of invisibly quantitative features, radiomics nomogram based on the whole tumor could be valuable in differentiating SNMM from sinonasal lymphoma with satisfactory efficiency. The nomogram outperformed both the new and the experienced radiologist, which is different from previous radiomics studies. This demonstrates its special value on rarities. In addition, the individualized 3D nomogram showed slight better performance than the individualized 2D nomogram or the generalized 3D nomogram based on only quantitative radiomics features. It indicated that the whole-volume-based analysis was suggested for future radiomics studies,(3) and fitting the inextractable quality features to the radiomics-based nomogram was more appropriate.

Conclusions

This was the first study to use MRI radiomics-based nomogram for preoperative prediction of SNMM, and it could be helpful to differentiate SNMM from sinonasal lymphoma.

Acknowledgements

This work was supported by the Grant of Science and Technology Commission of Shanghai Municipality (Grant number: 17411962100) and Key Project of the National Natural Science Foundation of China (Grant number: 61731009).

References

1. Lopez F, Rodrigo JP, Cardesa A, et al. Update on primary head and neck mucosal melanoma. Head Neck 2016;38(1):147-155. 2. Betts AM, Cornelius R. Magnetic resonance imaging in sinonasal disease. Top Magn Reson Imaging 2015;24(1):15-22. 3. Liu Y, Zhang Y, Cheng R, et al. Radiomics analysis of apparent diffusion coefficient in cervical cancer: A preliminary study on histological grade evaluation. J Magn Reson Imaging 2018.