Endocrine Tumor Classification via Machine-Learning-Based Elastography: A Systematic Scoping Review.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Elastography maps tissue stiffness, aiding in endocrine tumor detection.
• Machine learning (ML) enhances the accuracy and reliability of diagnostic imaging.
• Integrating ML with elastography offers improved diagnostic capabilities for endocrine tumors.

Purpose of the Study:

• To review the applications and performance of ML-based elastography for endocrine tumor classification.
• To synthesize findings from recent studies on ML and elastography in endocrine oncology.

Main Methods:

• A systematic literature search was conducted across major electronic databases.
• Eleven eligible articles focusing on thyroid and pancreatic tumors were reviewed.
• Methods included shear-wave and strain elastography, traditional ML, and deep learning (CNN, MLP, LSTM).

Main Results:

• All reviewed ML-elastography methods achieved diagnostic accuracy of 80% or higher.
• The highest accuracy for thyroid tumors was 94.70% using sequential training CNN.
• The highest accuracy for pancreatic tumors was 98.26% using a CNN-LSTM model integrating multiple imaging data.

Conclusions:

• Machine learning-based elastography is a highly effective tool for classifying endocrine tumors.
• Advanced deep learning models, particularly CNNs and LSTMs, demonstrate superior performance.
• Integrating elastography with other imaging modalities and ML techniques significantly improves diagnostic accuracy.