Automatic differentiation of thyroid scintigram by deep convolutional neural network: a dual center study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Nuclear Medicine

Background:

• Thyroid scintigraphy using 99mTc-pertechnetate is crucial for evaluating thyroid disease.
• Interpretation of thyroid scintigrams shows moderate physician consistency.
• Automated classification of thyroid scintigram patterns is needed.

Purpose of the Study:

• To develop an artificial intelligence (AI) system for automatic classification of four thyroid scintigram patterns.
• To evaluate the performance of different AI models in classifying thyroid scintigrams.

Main Methods:

• Collected 3087 thyroid scintigrams for training and validation.
• Implemented four pre-trained neural networks (ResNet50, DenseNet169, InceptionV3, InceptionResNetV2) using transfer learning.
• Evaluated models using accuracy, sensitivity, specificity, PPV, NPV, recall, precision, and F1-score.

Main Results:

• All four AI models exceeded 90% accuracy in classifying thyroid scintigrams.
• InceptionV3 demonstrated the highest performance with 92.73% internal and 87.75% external validation accuracy.
• Area under the curve (AUC) values for InceptionV3 were high across all four patterns in both internal and external validation.

Conclusions:

• Deep convolutional neural network-based AI models show significant performance in classifying thyroid scintigrams.
• AI systems can potentially improve the consistency and efficiency of thyroid scintigram interpretation by physicians.