Solid parotid tumors: an individual and integrative analysis of various ultrasonographic criteria. A prospective and observational study.

OBJECTIVES: The purpose of the study is to identify and validate ultrasound criteria for parotid tumors evaluation, as well as to elaborate a multimodal, multi-criteria and integrative ultrasound approach for allowing tumor discrimination in a non-invasive manner. MATERIAL AND METHOD: Twenty patients with solid parotid tumors (12 benign, 8 malignant) were examined by ultrasound: real-time "grey scale" ultrasound, Doppler ultrasound, elastography, harmonic ultrasound imaging with i.v. contrast (CEUS). The study focused on tumor morphology and circulation. The analysis of the results was observational, enhanced by statistical methods and artificial intelligence (decision trees). RESULTS: All malignant tumors showed increased hypoechogenicity, tumoral cervical adenopathies, increased stiffness and "in block" mobility with the parotid gland upon palpation with the transducer, uneven distribution of the contrast material during the arterial phase (8/8). To varying degrees, they showed imprecise delineation (7/8), structural heterogeneity (6/8) and disorganized flow pattern (6/8). All cases of benign tumors showed heterogeneous echostructure, clear delineation and no capsule (12). They also showed moderate hypoechogenicity (9/12), no cervical lymph nodes (11/12) and variable rigidity (increased 6/12; low 3/12). A selection and ranking of relevant ultrasound parameters was also made. Some of them were included in a transparent and easy-to-use decision tree model with 100% data accuracy. CONCLUSIONS: The characterization and discrimination of solid parotid tumors require a multimodal and multicriteria approach. Ultrasound criteria can be divided into criteria of certainty and criteria of diagnosis probability. CEUS examination of parotid tumors did not reveal significant differences between benign and malignant circulatory bed. Decision trees discovered by artificial intelligence from the data may represent intelligent diagnosis support systems with very high accuracy, up to 100%.