Keywords: thyroid malignancy, artificial intelligence, smart thyroid ultrasound software, ultrasound screening, diffuse or focal thyroid pathology
This project has three stages. The first step was the development of a smart computerized diagnostic algorithm used to stratify the risk in thyroid pathology, Ultrasound-based. It set the optimum time to achieve a thyroid biopsy (FNAB). We have used the latest international classifications (two-international-scores: EU-TIRADS/ACR_TIRADS), besides a scoring made by us, correlated with the pathological-results. The second stage included a Targeted Thyroid Screening in a population with high-risk, statistically significant. Finally, we are launching a cross border interdisciplinary multicenter US Screening.
How can we early diagnose thyroid malignancies at the high-risk population in primary healthcare by using new medical technology and artificial intelligence?
We report a targeted thyroid screening performed on 4386 - apparently healthy - adults with oncological risk factors+, aged over 20 years, followed for five years. We used the TIRADS classification by Russ modified and Strain Elastography, with both the elastographic scores by Rago and semiquantitative Strain Ratio (SR), for standardization and to establish if fine needle aspiration biopsy (FNAB) should be performed. The positive patients with focal thyroid lesions founded at this screening by family doctors were validated by endocrinologists through ultrasonography, FNAC, and histopathological or cytological examination. We designed an Ultrasound Scoring System (USS) for predicting malignancy and a diagnostic algorithm software. All patients were stored and counted into a Smart Thyroid Ultrasound Software. Finally, we compared ultrasound scores designed by us, with the histological results as Gold Standard method.
In this study they were found: 861 patients with thyroid diffuse disease and 696 with focal lesions. Prevalence of thyroid pathology was: 38.99% (95% CI:37.54% to 40.45%) with screening sensitivity: 96.49% specificity: 96.52% and a high accuracy of 96.51%, PPV:94.66%, NPV:97.73%, statistically significant, p<0.01. The ROC analysis of our US methods confirmed a higher level of diagnostic accuracy of Strain Elastography, p<0.001, AUC=0,995, 95% CI:0,97 to 1. Our cut-off-value of SR was: 2.5.
Performing Doppler Triplex Ultrasound Screening together with Strain Elastography, had the best accuracy for the analysis of the vascular network and the tumor stiffness, for differentiating 'benign versus malignant' of the thyroid tumors and for diagnosis of the diffuse thyroid diseases by family physicians with uses of the artificial intelligence as a support tool for the risk stratification.
Points for discussion:
How can we improve the early diagnosis of thyroid malignancies in the context of increasing their prevalence in industrialized countries?
Is it possible to perform thyroid ultrasound in a multidisciplinary screening team by family doctors specially trained in this regard?
How can artificial intelligence help us with ultrasonography technology as a diagnostic method in the practice of family doctors?