USING AI IN THE DIAGNOSIS OF THYROID DISEASE: A SINGLE CENTRE STUDY

AUTHORS:
O. Lozan (Mainz, Germany) , J. Staubitz-Vernazza (Mainz, Germany) , N. Bouzakri (Mainz, Germany) , R. Margies (Mainz, Germany) , A. Lederer (Mainz, Germany) , T.J. Musholt (Mainz, Germany)
Background:
Ultrasound is the most important method to assess thyroid nodules, which are classified according to TIRADS. However, it can be challenging for younger surgeons and endocrinologists alike. Artificial intelligence can be a tool to support examiners to improve the accuracy.
Methods:
Between March 2023 and December 2023, we examined 69 patients aged from 11 to 82 years old using the PIUR tUS Infinity Software. As a result, 107 nodules were scanned and analysed based on the ACR-TIRADS classification.
Results:
Out of 107 nodules, 70,09% (n=75) were classified the same as an experienced surgeon. 29,91% (n=32) were classified differently. 37,68% of the 69 patients (n=26) underwent surgery to remove the nodules (Hemithyroidectomy or Thyroidectomy) 15,94% (n=11) are scheduled for surgery 46,38% (n=32) patients had no indications for surgery but will be regularly examined in the outpatient clinic. Out of the 26 surgeries, 76,92% (n=20) were benign thyroids nodules, 7,69% (n=2) were medullary carcinomas (classified as TIRADS 4 and 5 respectively), and 15,38% (n=4) papillary carcinomas (classified as 2, 4, 4 and 5 respectively by the AI, but 4 - 5 by the experienced surgeon).
Conclusions:
In order to determine the functionality of the AI more data should be collected. In the preliminary testing, MTC are easier recognizable as malignant than PTC by the AI AI is a good tool in diagnosing thyroid lesions but cannot to this date replace an experienced surgeon. The software was easy to use, but required a significantly increased amount of time.