Fig 1.
Papillary thyroid carcinoma two-dimensional ultrasound image a-e and color Doppler flow image f.
a The aspect ratio of the nodule is taller than high. b The nodule is irregular in shape and has an unclear boundary with the thyroid capsule (arrows). c Microcalcification is seen within the nodule (arrows). d Burrs and horns are seen on the edge of the nodule (arrows). e Multiple nodules are found and fused into a mass (arrows). f Rich blood flow signals are seen within the nodule (arrows).
Fig 2.
Two-dimensional ultrasound images a-e and color Doppler flow images f of cervical lymph node metastasis in papillary thyroid carcinoma.
a Microcalcification was found in lymph node (arrows).b lymphatic portal misalignment in lymph node (arrows), and the boundary between cortex and medulla was unclear (arrows). c partial cystic degeneration was found in lymph node (arrows).d hyperechoic mass was found in lymph node (arrows).e multiple lymph node fusions were found in lymph node (arrows).f blood flow signal was found at the edge of lymph node (arrows).
Fig 3.
a Clinical model establishment process. b Clinical-Rad model establishment process. c deep learning model establishment process and ViT principle diagram. d model-assisted ultrasound doctor diagnostic process.
Fig 4.
Conclusion and exclusion criteria for patients.
Table 1.
Performance Comparison of Clinical Model, Clinical-Rad Model, and ViT.
Table 2.
The performance comparison of different deep learning models.
Fig 5.
The AUC of the ViT model, clinical model and combined radiomics model.
a, b and c respectively refer to the training set, internal validation queue and external validation queue. d Delong test results of the ViT model, clinical model and combined radiomics model. e Decision curve analysis of the ViT model, Clinical model and Clinical-Rad model.
Table 3.
Compare the diagnostic performance between ultrasound doctors with and without the assistance of the ViT model.
Fig 6.
Performance comparison between radiologists with and without ViT assistance in the internal testing cohort(a) and external-testing cohort(b).
Doctor C junior experience. Doctor D senior experience.