Introduction
Salivary gland tumours are rare and constitute approximately 3-6% of all head and neck neoplasms. Most lesions are benign and located in the parotid gland. The preoperative differentiation between benign and malignant lesions is of crucial importance for patient management. However, the radiological findings, as well as fine needle aspiration cytology, may considerably overlap. This study aims to develop and evaluate a deep learning network for characterizing salivary gland tumours, based on non-contrast CT images.
Materials and methods
Pre-operative non-contrast CT volumes of 88 salivary gland tumours were included. After the 3D segmentation of the lesion with a modified V-Net model, ResNet-50 was trained to classify benign and malignant lesions as selected 2D slices of the region. The training and test sets consisted of 61 and 27 cases, respectively.
Results
Regarding the segmentation step, our methodology obtained on the test set the Dice score 0.85, and the 95% quantile-Hausdorff distance 4.6 on average. For the final step of the classification, the obtained accuracy is 0.89 and the F1-score 0.88 on average.
Conclusion
The proposed model has proven promising for salivary gland tumour characterization. This might improve patient management and aid the surgical strategy making a more accurate preoperative lesion classification.

**Introduction**
Salivary gland tumours are rare and constitute approximately 3-6% of all head and neck neoplasms. Most lesions are benign and located in the parotid gland. The preoperative differentiation between benign and malignant lesions is of crucial importance for patient management. However, the radiological findings, as well as fine needle aspiration cytology, may considerably overlap. This study aims to develop and evaluate a deep learning network for characterizing salivary gland tumours, based on non-contrast CT images.
**Materials and methods**
Pre-operative non-contrast CT volumes of 88 salivary gland tumours were included. After the 3D segmentation of the lesion with a modified V-Net model, ResNet-50 was trained to classify benign and malignant lesions as selected 2D slices of the region. The training and test sets consisted of 61 and 27 cases, respectively.
**Results**
Regarding the segmentation step, our methodology obtained on the test set the Dice score 0.85, and the 95% quantile-Hausdorff distance 4.6 on average. For the final step of the classification, the obtained accuracy is 0.89 and the F1-score 0.88 on average.
**Conclusion**
The proposed model has proven promising for salivary gland tumour characterization. This might improve patient management and aid the surgical strategy making a more accurate preoperative lesion classification.

Deep learning for salivary gland tumours segmentation and classification based on CT images

[Networking hour](https://app.wonder.me/?spaceId=fe093845-759f-448f-9f55-01d3adccd820)
[COIs](https://drive.google.com/drive/folders/1nbfwAa9lf9Kz-wAZ7iLKhfpBFjpN_gkn?usp=sharing)