MEDO-Thyroid

MEDO-Thyroid is a software medical device employing machine learning to analyze thyroid ultrasound images, enabling automated quantification of thyroid lobe and nodule volumes to support radiologists in their diagnostic workflow. The device aims to provide consistent and accurate volumetric measurements from DICOM images, with performance validated across various ultrasound systems, thereby aiding in the evaluation and documentation of thyroid abnormalities.

MEDO-Thyroid is a cloud-based software as a medical device (SaMD) designed to assist qualified medical professionals in the assessment of thyroid ultrasound images for adult patients (18 years and older). Its primary purpose is to enhance the radiologist's workflow by facilitating the evaluation, quantification, and reporting of thyroid gland characteristics. The device utilizes machine learning techniques to analyze DICOM-format ultrasound images, enabling users to view and quantify thyroid lobes and identify, measure, and document thyroid nodules. Key functionalities include single and multi-frame image visualization, manual and semi-automatic landmark placement for measurements, and the generation of examination reports, which can incorporate TI-RADS classification based on user input. The software provides tools for measuring the volume of thyroid lobes and nodules, supporting both single and multi-frame ultrasound images. Performance data demonstrates its quantitative capabilities, with thyroid lobe volume measurements showing an overall Intraclass Correlation Coefficient (ICC) of 0.972 and a maximum percent volume error of 21.7% when compared to reference data. For thyroid nodule volume measurements, the overall ICC was 0.973 with a maximum percent volume error of 22.9%. The system was tested using images from Philips, GE, and Siemens ultrasound machines with high-frequency linear transducers, and its performance for nodule measurement was found to be independent of nodule size within the tested range of 0.13 cc to 36.5 cc.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the machine learning algorithm in MEDO-Thyroid differentiate between thyroid tissue, nodules, and surrounding structures, and what are its specific sensitivities and specificities for nodule detection beyond volume measurement?; What was the methodology for the "reference data" used to establish ground truth for thyroid lobe and nodule volume measurements in the performance testing, and what was its inter-observer variability?; While TI-RADS classification is based on user manual input, does the MEDO-Thyroid software offer any image analysis features that could aid or guide the user in selecting appropriate TI-RADS descriptors?; Given the stated maximum percent volume errors (around 21-23%), how might this level of measurement variability impact clinical decision-making, particularly for small nodule follow-up or growth assessment?; What is the performance of the semi-automatic landmark placement feature in terms of accuracy and time-saving compared to fully manual placement, and how does user expertise affect this performance?