Re-Modelling 99m-Technetium Pertechnetate Thyroid Uptake; Statistical, Machine Learning and Deep Learning Approaches.

Area of Science:

• Nuclear medicine
• Medical imaging
• Artificial intelligence

Background:

• Thyroid function assessment using 99mTc uptake has limitations.
• AI in nuclear medicine necessitates re-evaluating thyroid functional assessment benchmarks.

Purpose of the Study:

• To compare AI algorithms with conventional methods for thyroid function assessment.
• To evaluate the accuracy of machine learning and deep learning in thyroid scintigraphy.

Main Methods:

• Retrospective analysis of 123 patients comparing scintigraphy to biochemistry.
• Utilized conventional statistics, artificial neural networks (ANN), and convolutional neural networks (CNN).

Main Results:

• Machine learning (ML) achieved 84.6% accuracy with biochemistry features.
• Deep learning (DL) achieved 80.5% accuracy using image data alone.
• Thyroid uptake classification at 4.5% showed 82.6% accuracy for hyperthyroidism.

Conclusions:

• Thyroid scintigraphy with a validated cutoff (4.5%) aids in identifying hyperthyroid patients for radioiodine therapy.
• ML/DL algorithms can serve as second readers to improve accuracy and physician confidence.
• AI does not replace physicians but can augment their diagnostic capabilities.