Thy-Wise: An interpretable machine learning model for the evaluation of thyroid nodules

Data Description:
The database contains ultrasound images of thyroid nodules that were finally included in the study. As the aim of this study was to identify nodules as benign or malignant, all nodules were placed in two zip files according to their pathological nature: benign_after.zip and malignant_after.zip.
After unzipping the zip package and opening the folder, you can see several folders named by "pathological nature + number", each folder corresponds to a thyroid nodule and contains its ultrasound images collected in a single examination.

Ethical Approval:
This retrospective study was approved by the institutional Ethics Committees of the First Affiliated Hospital of Jinan University, and the requirement for informed consent was waived.

Sensitive Information Protection:
All sensitive information contained in the image, including the patient's personal information, the hospital visited, and the time of the visit, has been removed using the CV2 toolkit from python for the purpose of anonymization.

Processing pipeline and analysis steps:
All the annotations in the images and clips were eliminated before review. US images were evaluated in a blinded fashion, with no US or pathology reports available, by two board-certified radiologists (with more than 10 years of experience in thyroid sonography) independently.
Nodule size was measured as the maximal dimension on US images and the five gray-scale US categories were reviewed according to the ACR TI-RADS lexicon (5): composition, echogenicity, shape, margin, and echogenic foci. In the ACR TI-RADS, the TI-RADS risk level for nodules was determined by the total score of the five US categories, ranging from TR1 (benign) to TR5 (highly suspicious).