Enhanced thyroid nodule segmentation through U-Net and VGG16 fusion with feature engineering: A comprehensive study

Area of Science:

• Endocrinology and Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Thermography offers non-invasive thyroid imaging for early detection and risk stratification.
• Artificial intelligence (AI) shows promise in advancing medical diagnostics, particularly in thermal imaging analysis.
• Thyroid gland dysfunction impacts numerous bodily functions, necessitating improved diagnostic tools.

Purpose of the Study:

• To explore the potential of AI, specifically convolutional neural networks (CNNs), in analyzing thyroid thermograms.
• To enhance the detection of thyroid nodules and abnormalities using AI-driven thermal image analysis.
• To investigate AI's role in improving diagnostic accuracy for thyroid conditions.

Main Methods:

• Integration of AI and machine learning techniques for enhanced thyroid thermal image analysis.
• Proposed fusion of U-Net and VGG16 models combined with feature engineering (FE) for thyroid nodule segmentation.
• Leveraging feature engineering in transfer learning for nodule segmentation, even with limited datasets.

Main Results:

• Demonstrated efficacy of the AI approach across four studies, even with limited data.
• Significant improvement in dice coefficient observed with feature engineering (FE) in study 4.
• Incorporating radiomics with FE yielded substantial improvements in segmentation accuracy for small masked regions.

Conclusions:

• AI demonstrates significant potential for precise and efficient thyroid nodule segmentation.
• The proposed AI methods pave the way for enhanced thyroid health assessment.
• Further refinement of AI models can lead to even higher diagnostic accuracy.